The Virtual Battlefield: Perspectives On Cyber Warfare

THE VIRTUAL BATTLEFIELD: PERSPECTIVES
ON CYBER WARFARE
Cryptology and Information Security Series
The Cryptology & Information Security Series (CISS) presents the latest research results in the
theory and practice, analysis and design, implementation, application and experience of
cryptology and information security techniques. It covers all aspects of cryptology and
information security for an audience of information security researchers with specialized
technical backgrounds.
Coordinating Series Editors: Raphael C.-W. Phan and Jianying Zhou
Series editors
Feng Bao, Institute for Infocomm Research, Singapore Nasir Memon, Polytech University, USA
Kefei Chen, Shanghai Jiaotong University, China Chris Mitchell, RHUL, United Kingdom
Robert Deng, SMU, Singapore David Naccache, École Normale Supérieure, France
Yevgeniy Dodis, New York University, USA Gregory Neven, IBM Research, Switzerland
Dieter Gollmann, TU Hamburg-Harburg, Germany Phong Nguyen, CNRS / École Normale Supérieure, France
Markus Jakobsson, Indiana University, USA Andrew Odlyzko, University of Minnesota, USA
Marc Joye, Thomson R&D, France Adam Young, MITRE Corporation, USA
Javier Lopez, University of Malaga, Spain Moti Yung, Columbia University, USA
Volume 3
Recently published in this series
Vol. 2. M. Joye and G. Neven (Eds.), Identity-Based Cryptography

Vol. 1. J. Lopez and J. Zhou (Eds.), Wireless Sensor Network Security
ISSN 1871-6431
The Virtual Battlefield: Perspectives
on Cyber Warfare
Edited by
Christian Czosseck
Cooperative Cyber Defence Centre of Excellence (CCD CoE), Tallinn, Estonia
and
Kenneth Geers
Naval Criminal Investigative Service (NCIS), and Cooperative Cyber
Defence Centre of Excellence (CCD CoE), Tallinn, Estonia
Amsterdam • Berlin • Tokyo • Washington, DC

© 2009 The authors and IOS Press.
All rights reserved. No part of this book may be reproduced, stored in a retrieval system,
or transmitted, in any form or by any means, without prior written permission from the publisher.
ISBN 978-1-60750-060-5
Library of Congress Control Number: 2009937744
Publisher
IOS Press BV
Nieuwe Hemweg 6B
1013 BG Amsterdam
The Netherlands
fax: +31 20 687 0019
e-mail: order@iospress.nl
Distributor in the USA and Canada

IOS Press, Inc.
4502 Rachael Manor Drive
Fairfax, VA 22032
USA
fax: +1 703 323 3668
e-mail: iosbooks@iospress.com
LEGAL NOTICE
The publisher is not responsible for the use which might be made of the following information.
PRINTED IN THE NETHERLANDS

The Virtual Battlefield: Perspectives on Cyber Warfare v
C. Czosseck and K. Geers (Eds.)
IOS Press, 2009
© 2009 The authors and IOS Press. All rights reserved.
Preface
On January 14, 2009, I posted a Call for Papers (CFP) to Bugtraq for a Conference on
Cyber Warfare. Within hours, I received an email from n3td3v, an infamous computer
security commentator [1]:
How can you have a security conference on “cyber warfare” when

it doesn’t exist and has never taken place.
n3td3v has a point. Estimating the threat posed by cyber attacks is not easy. Case stud-
ies are few in number, much information lies outside the public domain, and there have
been no wars – yet – between modern, cyber-capable militaries. While the era of cyber
espionage is already here [2], a possible era of broad-scale cyber warfare still lies in the
future [3].
Nevertheless, an examination of international affairs over the past two decades
suggests that cyber battles of great consequence are easy to find. Since the earliest days
of the World Wide Web, Chechen guerilla fighters, armed not only with rifles but with
digital cameras and HTML, have clearly demonstrated the power of Internet-enabled
propaganda. During the 1999 war over Kosovo, likely non-state actors tried to disrupt
NATO military operations through computer hacking, and were able to claim minor
victories [4]. In 2007, Syrian air defense was reportedly disabled by a cyber attack
moments before the Israeli air force demolished an alleged Syrian nuclear reactor [5].
In 2009, the entire nation-state of Kyrgyzstan was knocked offline during a time of
political crisis [6].
What military officers call the ‘battlespace’ grows more difficult to define – and to
defend – over time. Advances in technology are normally evolutionary, but they can be
revolutionary: artillery reached over the front lines of battle; rockets and airplanes
crossed national boundaries; and today, cyber attacks can target political leadership,
military systems, and average citizens anywhere in the world, during peacetime or war,
with the added benefit of attacker anonymity.
Information Technology (IT) now pervades our lives. In 1965, Gordon Moore cor-
rectly predicted that the number of transistors on a computer chip would double every
two years [7]. There has been concomitant growth in almost all aspects of IT, including
the widespread availability of practical encryption, user-friendly hacker tools, and
Web-enabled open source intelligence (OSINT). It should therefore no longer be sur-
prising that political and military strategists use and abuse computers, databases, and
the networks that connect them in order to achieve their objectives [8]. In the early
1980s, this concept was already known in the Soviet Union as the Military Technologi-
cal Revolution (MTR); following the U.S. victory in the 1991 Gulf War, the Penta-
gon’s Revolution in Military Affairs was almost a household term [9].
Cyberspace, narrowly defined, is a collection of networked computers. But the ex-
tent to which humans (and other computers) obtain their information and marching
orders from somewhere in cyberspace grows by the day. This is in part what hackers
call the expanding ‘attack surface’. In national security terms, the concepts of attack,
defense, and security remain unchanged, as do the threats posed by adversary propa-
vi
ganda, espionage, and attacks on critical infrastructure. The difference is that tradi-
tional threats are now Internet-enabled; they employ a new delivery mechanism that
can increase the speed, diffusion, and even the power of an attack. The cyber skir-
mishes we witness today likely foreshadow a long march that cyber warfare will make
from a corollary of real-world disputes to a lead role in conflicts of the future.
This book consists of the research papers presented at the Cooperative Cyber De-
fense Centre of Excellence (CCD CoE) Conference on Cyber Warfare, which took
place in Tallinn, Estonia, in June 2009. Individually and collectively, they explore the
relationship between computer security and national security. Unsurprisingly, the devil
is found in the details: the challenge of attribution, the calculation of damages, the se-
curity of critical infrastructure, ethics, jurisdiction, responsibility, and much more. This
book is divided into two sections: Strategic Viewpoints and Technical Challenges and
Solutions.
Strategic Viewpoints
Chapter 1: “Cyber Wars: A paradigm shift from Means to Ends.” Amit Sharma, from
the Indian Ministry of Defence, argues that cyber warfare is different from other types
of conflict, and merits its own set of rules. He warns against trying to fit cyber warfare
into the traditional definitions found in the Law of Armed Conflict (LOAC). Sharma
believes that it is easy to underestimate the strategic potential of cyber warfare, stating
that cyber attacks alone are powerful enough to achieve political goals. In his view,
cyber warfare will cease to be merely a force multiplier for conventional warfare;
rather, conventional warfare will be used to support the objectives of cyber warfare.
Sharma examines the legal treatment of nuclear weapons, and deterrence theory, for
possible application in the cyber domain.
Chapter 2: “Towards an Evolving Theory of Cyberpower.” Dr. Stuart H. Starr,
from the Center for Technology and National Security Policy (CTNSP) at the National
Defense University (NDU), discusses the development of a theory of ‘cyberpower’ in
research that systematically addresses five key areas: it defines the key terms that are
associated with cyber issues; it categorizes the elements, constituent parts, and factors
that yield a framework for thinking about cyberpower; it explains the major factors that
are driving the evolution of cyberspace and cyberpower; it connects the various ele-
ments of cyberstrategy so that a policy maker can place issues in proper context; and it
anticipates key changes in cyberspace that are likely to affect decision making.
Chapter 3: “Sub Rosa Cyber War.” Martin C. Libicki of the RAND Corporation
explains that the battlefield terrain of cyberspace allows not only for stealthy attack and
defense, but even for the existence of a stealthy war. Sub rosa cyber war is a conflict in
which the warring parties do not publicly acknowledge battlefield victories and defeats,
or even the existence of an ongoing war. Two reasons such a conflict is possible in-
clude the difficulty of good cyber battle damage assessment and the diabolical chal-
lenge of cyber attack attribution. Further, opponents may desire to keep a cyber conflict
sub rosa in order to preserve freedom of action; public awareness and scrutiny could
complicate negotiations or lead to unwanted escalation. Libicki cautions that sub rosa
cyber war carries serious risks, such as insufficient operational oversight and a dubious
assumption that the conflict is truly sub rosa to third parties.
Chapter 4: “Warfare and the Continuum of Cyber Risks: A Policy Perspective.”
Andrew Cutts, the Director of Cyber Security Policy at the U.S. Department of Home-
vii
land Security, explains that nation-states are beginning to appreciate the potential bene-
fits and costs of employing cyber attacks as a means of projecting national power. He
offers a framework for evaluating national cyber security policy that includes prioritiz-
ing competing missions and balancing short- and long-term objectives. The focus of
this chapter is on the long-term, strategic threat. To get ahead of the most serious cyber
risks, national security leadership must strive to find the appropriate balance of re-
sources, energy, and focus, specifically to distinguish between the most frequent threats
and those that are the most consequential.
Chapter 5: “Cyber Terrorism: A New Dimension in Battlespace.” Major J. P. I. A.
G. Charvat, from the Centre of Excellence Defence against Terrorism in Ankara, Tur-
key, examines the emerging threat of cyber terrorism. First, he discusses the phenome-
non of terrorism and the motivations of terrorist organizations. Next, he explores the
way terrorists now use IT to disseminate propaganda, to recruit new members, and to
support conventional attacks. Finally, he considers the possibility that a terrorist or-
ganization might adopt a pure cyber attack strategy in an attempt to inflict electronic or
physical damage.
Chapter 6: “Borders in Cyberspace: Can Sovereignty Adapt to the Challenges of
Cyber Security?” Forrest Hare, from the School of Public Policy at George Mason
University, writes that while cyberspace has no borders, nation-states do. Therefore,
governments must consider how they will define and defend their sovereignty in this
new domain. Crucially, Hare argues that governments should realize that to some de-
gree they will be forced to coordinate and integrate their efforts on the international
level. To help explore the nature of boundaries in cyberspace, this research makes use
of two very different frameworks: the first is a comparison of the challenges of cyber
security to international drug trafficking; the second employs game theory to support
the cyber security decision-making process.
Chapter 7: “Towards a Global Regime for Cyber Warfare.” Dr. Rex Hughes, Co-
Director of the Cyber Security Project, Chatham House, London, explains that when
novel, disruptive technologies dramatically alter the nature of warfare, history shows
that the likelihood of a new arms race is high. As more nations aspire to project na-
tional power in cyberspace, a digital arms race may be around the corner. Therefore,
diplomats should find a coherent set of principles, rules, and norms to govern state se-
curity and military operations in cyberspace. Hughes argues that the most important
cyber challenge facing national security thinkers today concerns how to prevent a ma-
jor arms race in this arena. This chapter introduces readers to the Law of Armed Con-
flict, examines how it might apply to cyber warfare, and outlines the steps required to
create a global regime for cyber warfare.
Chapter 8: “What Analogies Can Tell Us About the Future of Cybersecurity.”
David Sulek and Ned Moran of Booz Allen Hamilton examine the benefits and draw-
backs of using historical analogies to understand cyber warfare. When such analogies
are appropriately chosen and systematically applied, they can clarify the present situa-
tion and offer decision-makers strategic insight; vice versa, poor analogies obscure
objectives, unnecessarily complicate choices, and create blind spots. In every case,
analogies are bound to fail unless they incorporate objective analysis and their hand is
not overplayed. The authors consider the well-known Electronic Pearl Harbor, and ex-
plore a range of new ideas: Cyber Katrina, Cyber Sputnik, Cyber Balkanization, Cyber
Tribes, Cyber Conquistadors, and Cybernization.
Chapter 9: “The Information Sphere Domain – Increasing Understanding and Co-
operation.” In this chapter, Dr. Patrick D. Allen (Johns Hopkins University, Applied
viii
Physics Lab) and Dennis P. Gilbert, Jr (Booz Allen Hamilton) write that a great advan-
tage always accrues to a competitor who understands and operates within a domain
better than their opponent. First, the authors define what constitutes a domain, and de-
scribe how new domains are created over time. Their research outlines the ‘Information
Sphere’ domain, which has features that are both similar to and different from the four
traditional, physical domains: air, land, sea, and space.
Chapter 10: “Sun Tzu was a Hacker: An Examination of the Tactics and Opera-
tions from a Real World Cyber Attack.” Billy K. Rios, of GreyLogic, LLC, examines
the nuts and bolts of a real world, international cyber attack. Rios was in a unique posi-
tion to witness the “communications, execution, and responses” of the cyber attackers
and defenders in the 2008 war between Russia and Georgia. The author concludes that
the availability and effectiveness of cyber attack tools ensure that data packets, fired
purely on the battlefield terrain of computer networks, are likely to play a role in all
future international conflicts. This research paper considers cyber attacks in the light of
traditional concepts of maneuver warfare, as described in Marine Corps Doctrinal Pub-
lication 1 (MCDP-1).
Chapter 11: “Belarus in the Context of European Cyber Security.” Fyodor Pav-
lyuchenko, of www.charter97.org, examines the use of Internet censorship as a gov-
ernment tool that can be used to suppress political dissent within a nation-state. The
author, representing one of the most popular – and hacked – online news sites in Bela-
rus, catalogues a decade of cyber attacks that the site has suffered (usually during times
of political tension). The author believes that the use of political DoS attacks threatens
not only freedom of expression in Belarus, but the integrity of Internet resources in
other European countries as well. Finally, he contends that the cyber conflict in Belarus
is analogous to the ongoing struggle between state and non-state actors in Russia.
Chapter 12: “Politically Motivated Denial of Service Attacks.” Jose Nazario of
Arbor Networks dissects DDoS attacks that have been used in support of political or
ideological goals. He explains how DDoS attacks have evolved from a platform used to
inflict “punitive damage” on a target to a sophisticated means of Internet censorship.
Nazario uses extraordinary access to a wide range of data, including Internet backbone
traffic, border gateway protocol (BGP) routing data, botnet communications, and
community chatter, to create a compelling narrative. He concludes that while most
Internet attackers appear to be non-state actors, they are nonetheless capable of using
botnets of significant size and power to launch effective DDoS attacks.
Chapter 13: “A Brief Examination of Media Coverage of Cyberattacks (2007-
Present).” Cyrus Farivar, a freelance technology journalist, critiques the media cover-
age of three politically-oriented cyber attacks: Kyrgyzstan (2009), Georgia (2008), and
Estonia (2007). He reflects on where journalistic analysis was correct and where it was
off the mark. On balance, he argues that there is enormous room for improvement, and
that it is in the interest of the cyber security community, the media, and policymakers
to improve their understanding of and their ability to write on the subject of cyber at-
tacks, so that public understanding and appreciation of this new threat will increase.
Technical Challenges and Solutions
Chapter 14: “Behavioral Analysis of Zombie Armies.” Olivier Thonnard, Wim Mees
(both from the Royal Military Academy, Belgium) and Marc Dacier (Symantec
ix
Research Labs, France) present ground-breaking research on the behavior of ‘zombie

armies’. They characterize the long-term behavior, global characteristics, strategic evo-
lution, size, lifespan, and resilience of botnets. Their research highlights the uneven
spatial distribution of infected computers across the Internet, specifically on a limited
number of “unclean” or “zombie-friendly” networks. Most of the botnets they studied
have an impressive attack capability in terms of bandwidth and the number of ways
they are able to probe and exploit other computers; further, like real-world armies, they
can coordinate their efforts with other botnets. Their analysis employed data from the
European Union (EU)-funded WOMBAT project (Worldwide Observatory of Mali-
cious Behaviors and Attack Threats).
Chapter 15: “Proactive Botnet Countermeasures – An Offensive Approach.” Felix
Leder, Tillmann Werner, and Peter Martini, from the Institute of Computer Science IV,
University of Bonn, Germany, contend that cyber defenders are at a disadvantage vis-à-
vis attackers in part because computer science evolves too quickly for cyber law to
keep up. In this chapter, the authors describe a technically feasible, proactive way to
detect and defeat computer botnets, based on the assumption that reactive measures
alone are insufficient. Their research formalizes botnet topologies, via real-world ex-
amples, and derives effective strategies for attacking them. However, they explain that
their approach employs tactics that go well beyond what current cyber law encom-
passes, and argue that “controversial discussions” are needed to explore the political,
legal, ethical, and liability-based ramifications of a proactive, counter-botnet ap-
proach – sooner rather than later.
Chapter 16: “When Not to Pull the Plug – The Need for Network Counter-
Surveillance Operations.” Scott Knight and Sylvain Leblanc of the Royal Military Col-
lege of Canada argue that one traditional response to a cyber attack – to immediately
remove compromised machines from the network – has two primary drawbacks: it
warns the intruder that he or she has been discovered, and it can prevent the collection
of crucial evidence to help determine motive and estimate damage. The authors deline-
ate a Network Counter-Surveillance Operation designed to maintain quiet contact with
a hacker while information is gathered on intentions, techniques and more.
Chapter 17: “Autonomic Computer Network Defence Using Risk State and Rein-
forcement Learning.” Luc Beaudoin (Defense Research and Development Canada),
Nathalie Japkowicz and Stan Matwin (both from the University of Ottawa) write that
humans are simply not able to handle the complexity and speed of many forms of cyber
attack, and that it is essential to automate certain aspects of traffic analysis and attack
mitigation. They argue that a level of autonomic computer network defense can be
achieved using reinforcement learning and dynamic risk assessment, with a view to-
ward determining optimal action sequences (or policies) to recover from computer
network risk situations. In their view, this approach will benefit commercial network
management and security products by aiding in the selection of automatic mitigation
actions, as risk states are sensed.
Chapter 18: “Enhancing Graph-based Automated DoS Attack Response.” Gabriel
Klein, Marko Jahnke, Jens Tölle (all from FGAN-FKIE, Germany), and Peter Martini
(University of Bonn, Germany) argue that timely and appropriate responses to DoS
attacks are critical in both civilian and military settings. While intrusion detection sys-
tems (IDS) are capable of detecting DoS attacks, the growing sophistication and speed
of such attacks increase the need for automated countermeasures to support computer
network defense. Per force, it is necessary to quickly evaluate the potential effects of
x
proposed countermeasures on network resources. This chapter discusses GrADAR, an

intuitive, graph-based approach for automatically assessing the likely effects of DoS
countermeasures on a network. Further, it proposes an enhancement which takes into
account the effects of the workload on resource availability.
Chapter 19: “On nth Order Attacks.” Daniel Bilar, from the University of New Or-
leans, USA, explores a class of cyber attack designed to subvert “mission-sustaining
ancillary systems”. In technical terms, ancillary systems could be throughput control,
visualization environments, memory resource allocation, manufacturing, and the supply
chain; the purpose of such systems – and the real target of the attacker – is its political,
military, or economic mission. The attacker’s goal could be to disrupt power manage-
ment, logistics, elections, or even the social welfare of the target. Bilar discusses his-
torical, current and forward-looking examples, with special emphasis on attacks against
computerized, open societies.
Chapter 20: “Business and Social Evaluation of Denial of Service Attacks in View
of Scaling Economic Countermeasures.” Louis-Francois Pau, from the Copenhagen
Business School and Rotterdam School of Management, writes that DoS attacks not
only affect computer network resources; they can have a direct, negative impact on the
bottom line of a business in the real world. This chapter proposes a method to deter-
mine the direct and indirect costs associated with DoS attacks, which is a necessary
step in determining countermeasures aimed at legal- or policy-driven dissuasion, re-
taliation, compensation, and restoration. Dr. Pau’s method relies on time-preference
dynamics applied to monetary mass for the restoration of capabilities, on long-term
investments to rebuild capabilities, and on the usability level of capabilities after an
attack. A real-world example of a DoS attack on a corporate data centre is provided. In
conclusion, the author gives specific policy recommendations and suggests information
exchange requirements.
Chapter 21: “Virtual Plots, Real Revolution.” Roelof Temmingh (Paterva) and
Kenneth Geers (NCIS/CCD CoE) investigate whether computer botnets could evolve
from spam and Distributed Denial of Service (DDoS) generators to semantic creatures
that could voice opinions, arguments, and even threats via the Internet. Key to their
argument is the assumption that only a small percentage of information on the Web is
truly unique. In theory, a malicious actor could create a virtual population of fraudulent
identities from stolen and/or randomized biographies, pictures, and histories of Internet
activity, which could be used to support a criminal, political, military or terrorist
agenda. The increasingly impersonal nature of Internet communications will make
timely threat evaluation difficult.
Many thanks to all who were involved in organizing the 2009 Conference on Cyber
Warfare, and especially to Christian Czosseck, for his diligence in helping to edit this
book.
Kenneth Geers
Tallinn, Estonia
August 7, 2009
xi
References
[1] “Who is “n3td3v”?”, Hacker Factor Solutions, White Paper, Release 1.4.1, 12-October-2006,
www.hackerfactor.com/papers/who_is_n3td3v.pdf.
[2] “Espionage report: Merkel’s China visit marred by hacking allegations”. Spiegel Online, August 27,
2007, http://www.spiegel.de/international/world/0,1518,502169,00.html; Cody, E. “Chinese official ac-
cuses nations of hacking”. Washington Post, September 13, 2007, http://www.washingtonpost.com/wp-
dyn/content/article/2007/09/12/AR2007091200791_pf.html#.
[3] However, this is already a point on which reasonable people can disagree: “How robot drones revolu-
tionized the face of warfare”, Cable News Network (CNN), July 24, 2009. http://edition.cnn.com/2009/
WORLD/americas/07/23/wus.warfare.remote.uav/index.html.
[4] Geers, Kenneth. “Cyberspace and the Changing Nature of Warfare,” SC Magazine, August 27, 2008,
http://www.scmagazineus.com/Cyberspace-and-the-changing-nature-of-warfare/article/115929/.
[5] Fulghum, David A., Wall, Robert, and Butler, Amy. “Cyber-Combat’s First Shot,” Aviation Week &
Space Technology. November 26, 2007. Vol. 167, Iss. 21, p. 28.
[6] Keizer, Gregg. “Russian ‘cyber militia’ knocks Kyrgyzstan offline”, Computerworld, 01/28/2009,
www.networkworld.com/news/2009/012809-russian-cyber-militia-knocks-kyrgyzstan.html.
[7] “Moore’s Law”, Intel Corporation, www.intel.com/technology/mooreslaw/.
[8] Adams, James. “Virtual Defense”, Foreign Affairs, May/June 2001, 80, 3, p. 98.
[9] Mishra, Shitanshu. “Network Centric Warfare in the Context of ‘Operation Iraqi Freedom’” Strategic
Analysis, Vol. 27, No. 4, Oct-Dec 2003, Institute for Defence Studies and Analyses, http://www.idsa.
in/publications/strategic-analysis/2003/oct/Shitanshu.pdf.
This page intentionally left blank
xiii
About the Contributors

Editors
Christian Czosseck is a Scientist at the Cooperative Cyber Defence Centre of Ex-

cellence (CCD CoE) in Tallinn, Estonia, and the German representative there since
2008. He has been a soldier in the German Armed Forces (Bundeswehr) for more than
12 years. Christian graduated first in his class in Computer Science at the Universität
der Bundeswehr in Munich, and for the past six years he has held numerous Informa-
tion Assurance positions in the German military.
Kenneth Geers, Naval Criminal Investigative Service (NCIS), is a Scientist and
the U.S. representative to the CCD CoE. He has served as an intelligence analyst, a
French and Russian linguist, and computer programmer in support of strategic arms
control initiatives. Kenneth has been a student in six countries, earned the Defense
Language Institute Provost’s Award for Outstanding Scholastic Achievement, wrote
his Master’s thesis on nuclear proliferation at the Jackson School of International Stud-
ies, University of Washington, Seattle, and is a Certified Information Systems Security
Professional (CISSP).
Authors
Luc Beaudoin earned a degree in electrical engineering with honors from the
Royal Military College of Canada, a Masters in Business Administration from the Uni-
versity of Québec, and is now a MSc student in Systems Sciences at the University
of Ottawa. Luc served ten years in the Canadian Forces as a telecommunications offi-
cer, where he was the first Watch Officer at the Canadian Forces Network Operations
Centre. He joined the Defence Research and Development Canada- Network Informa-
tion Operations section in 2003, where he leads network security projects associated
with situational awareness, dynamic risk response, decision making and automated
defense.
Daniel Bilar is an Assistant Professor of Computer Science at the University of
New Orleans, and holds a PhD in Engineering Sciences from Dartmouth College.
Daniel was a founding member of the Institute for Security and Technology Studies at
Dartmouth, conducting counter-terrorism technology research for the U.S. Departments
of Justice and Homeland Security. Active research topics include the detection, classi-
fication and containment of highly-evolved malicious software, quantitative risk analy-
sis/management of networks, and the optimization of business and innovation proc-
esses.
J. P. I. A. G. Charvat graduated from Anglia Ruskin University with a degree in
Modern History in 1992. After a year at the Royal Military Academy Sandhurst, he
was commissioned into the Royal Military Police in 1993. He has served in the UK,
Cyprus, Germany, Bosnia, and in the U.S. at the Army Military Police School, Fort
Leonard Wood. After company command, he was posted to the Centre of Excellence
xiv
Defence against Terrorism in Ankara, Turkey. Julian also holds a Certificate in Terror-
ism Studies from St Andrew’s University.
Andrew Cutts joined the Department of Homeland Security in March 2008, in the
Office of Policy as Director for Cyber Security. Prior to joining DHS, he directed the
Cyber Conflict Research Institute at Norwich University. From 2002–2004, he was a
research program manager at the Institute for Security Technology Studies at Dart-
mouth College. He was in the private sector from 1995 to 2002, where he became Vice
President for Information Technology at a high-tech engineering services firm based in
Connecticut. Before that, he served as a Naval Intelligence Officer for nine years.
Cyrus Farivar is a freelance technology journalist and radio reporter/producer
from Oakland, California (USA). He has lived in Lyon (France), Saint-Louis (Senegal),
Melbourne (Australia), and Geneva (Switzerland). He is currently writing a book tenta-
tively called “The Internet of Elsewhere,” due out from Rutgers University Press in
2010. He reports for National Public Radio, The World (WGBH/PRI/BBC), Canadian
Broadcasting Corporation, and freelances for The Economist, Foreign Policy, Slate,
The New York Times, Popular Mechanics, and Wired.
Kenneth Geers, Naval Criminal Investigative Service (NCIS), is a Scientist and
the U.S. representative to the CCD CoE. He has served as an intelligence analyst, a
French and Russian linguist, and computer programmer in support of strategic arms
control initiatives. Kenneth has been a student in six countries, earned the Defense
Language Institute Provost’s Award for Outstanding Scholastic Achievement, wrote
his Master’s thesis on nuclear proliferation at the Jackson School of International Stud-
ies, University of Washington, Seattle, and is a Certified Information Systems Security
Professional (CISSP).
Dennis P. Gilbert, Jr. is a Cybersecurity Strategist and Information Operations
(IO) Practitioner with Booz Allen Hamilton. Mr. Gilbert was Vice President in an IT
company, and served in the U.S. Air Force for 21 years, where he held leadership posi-
tions in Information Assurance (IA), satellite communications, space control, and elec-
tronic warfare. He now serves as a trusted advisor to organizations within the U.S. De-
partment of Defense and Intelligence Community.
Forrest Hare is a Lieutenant Colonel in the United States, assigned to the Office
of the Secretary of Defence. Most recently, he was responsible for developing the
United States Air Force cyberspace strategy as part of the Military Strategy for Cyber-
space Operations. In addition, he has served in numerous information operations posi-
tions world-wide. Lt Col Hare is currently a PhD student at the George Mason School
of Public Policy, studying national security policy for cyberspace. He has taught Eco-
nomics and Geography at the United States Air Force Academy and at the University
of Maryland, Asian Division. He received his Bachelor of Science degree from the
United States Air Force Academy, and a Master of Arts from the University of Illinois.
He also conducted post-graduate studies at the University of Fribourg, Switzerland,
under a Swiss University Grant.
Rex Hughes is a Senior Resident Member at Wolfson College, Cambridge and
Research Associate of The Cambridge-MIT Institute. Dr. Hughes’ current research
examines the global governance of cyber security. With Dr. Paul Cornish, he estab-
lished the Cyber Security Project at Chatham House which published in March 2009
its first major report, Cyberspace and the National Security of the United Kingdom.
Prior to his Cambridge years Hughes founded and directed the world’s first multidisci-
plinary Internet Studies program at the University of Washington in Seattle, where in
partnership with IBM-Lotus he led the development of iEnvoy, the first secure diplo-
xv
mat-to-diplomat Internet communications platform deployed by the U.S. Department of

State.
Marko Jahnke is a senior researcher at the Research Institute for Communication,
Information Processing, and Ergonomics (FKIE) in Wachtberg, Germany. He has sev-
eral years of experience in applied information and communications security. He cur-
rently leads a multi-participant research project that focuses on intrusion detection in
tactical MANETs.
Gabriel Klein is a researcher at the Research Institute for Communication, Infor-
mation Processing, and Ergonomics (FKIE) in Wachtberg, Germany. His research fo-
cuses on intrusion detection in tactical MANETs. He is currently spending a 4-month
visit as a guest researcher at the Cooperative Cyber Defence Centre of Excellence
(CCD-CoE) in Tallinn, Estonia.
Scott Knight is an Associate Professor in the Department of Electrical and Com-
puter Engineering at the Royal Military College (RMC) of Canada. Dr. Knight was
appointed to the academic faculty at RMC in 2000 on retirement from 21 years of ser-
vice in the Canadian Air Force. He founded the RMC Computer Security Laboratory,
and continues to lead this research group in his present appointment.
Sylvain P. Leblanc is an Assistant Professor and PhD candidate in Software Engi-
neering with the Department of Electrical and Computer Engineering at the Royal Mili-
tary College of Canada. Sylvain (Sly) served for 23 years as a Signals Officer with the
Canadian Forces. He is a member of the RMC Computer Security Laboratory, and
works closely with the Defence Research and Development Centre – Ottawa. His spe-
cific research interests are in active network defence and computer network defence
processes.
Felix Leder is a PhD student at the University of Bonn. After working for Nokia,
he turned his attention fully to his favourite field of research: IT security. His current
research interests are botnet mitigation tactics and new methodologies for executable
file and malware analysis. A lot of his spare time is spent working on the Honeynet
Project.
Martin Libicki (PhD, U.C. Berkeley) is a senior management scientist at RAND,
focusing on the impact of IT on national security. He is the author of Conquest in Cy-
berspace: National Security and Information Warfare, Information Technology Stan-
dards: Quest for the Common Byte, Cyber-Deterrence and Cyber-War, What is Infor-
mation Warfare, The Mesh and the Net: Speculations on Armed Conflict in a Time of
Free Silicon, and Who Runs What in the Global Information Grid. His most recent as-
signments included helping to organize the Air Force for cyber-war, exploiting cell
phones in counter-insurgency, developing a post-9/11 IT strategy for the U.S. Depart-
ment of Justice, using biometrics for identity management, assessing DARPA’s Terror-
ist Information Awareness program, conducting information security analysis for the
FBI, and evaluating In-Q-Tel.
Ned Moran is an Adjunct Professor in the Computer Science Department at
Georgetown University, and also works as an Intelligence Analyst with the Terrorism
Research Center. One of his particular areas of interest is how modern terrorist groups
use the Internet to organize and operate. In this capacity, Mr. Moran conducts cyber
investigations and forensics on suspected terrorist websites. Ned received an M.A. in
Security Studies with honors from Georgetown University’s School of Foreign Service
where he focused on Technology and National Security Policy.
Jose Nazario is the manager for security research at Arbor Networks, responsible
for analyzing Internet security threats, reverse engineering malicious code, and soft-
xvi
ware development. Dr. Nazario develops security mechanisms for Arbor’s Peakflow
platforms via the Active Threat Feed (ATF). His research interests include large-scale
Internet trends such as reachability and topology measurement, Internet-scale events
such as DDoS attacks, botnets and worms, source code analysis, and data mining. He is
the author of Defense and Detection Strategies against Internet Worms and Secure Ar-
chitectures with OpenBSD and maintains WormBlog.com, a site devoted to worm de-
tection and defense research.
Louis-Francois Pau is Professor of Mobile communications and media at the Co-
penhagen business school and at the Rotterdam school of management. Previously, he
was CTO of L.M. Ericsson’s Network Systems division and CTO for Digital Equip-
ment/Hewlett Packard Europe. He has been on the faculties of the Danish Technical
University, Ecole Nationale Supérieure des Télécommunications (Paris), M.I.T., and
University of Tokyo. He is a Fellow of IEEE (USA), BCS (UK), JSPS (Japan).
Fyodor Pavlyuchenko was born in Minsk, Belarus in 1976. He graduated from
Belarus State University in 1999. Since 1997, he has been involved in a number of
high-profile Belarusian Internet projects. Since 2000, he has developed, maintained and
promoted www.charter97.org, the leading independent news website in Belarus. Fyo-
dor created the IT-resource bybanner.com for the dissemination of information on Bel-
arusian Internet news, and has participated in the creation and development of Belaru-
sian newspapers and online publications.
Billy K. Rios is a Security Engineer for Microsoft, where he studies emerging
risks and cutting edge attacks and defenses. Previously, Billy was a Senior Security
Consultant for VeriSign, a penetration tester for the Ernst and Young Advanced Secu-
rity Center, a U.S. Department of Defense Intrusion Detection Analyst and Marine
Corps Officer. He has presented research at Blackhat, RSA, Bluehat, DEFCON,
PacSec, HITB, and ASIA.
Amit Sharma is Deputy Director/Scientist C in the Institute for System Studies
and Analysis, Defence Research and Development Organization, Ministry of Defence,
Government of India. He has worked in the field of Information Security, Strategic
Information Dissemination Systems, Net Centric Warfare, C4I2SR systems and Secure
and survivable networks. He did his B Tech (honors) in Computer Science and Tech-
nology at the National Institute of Technology, Hamirpur India, and is currently pursu-
ing his Masters in Global Security from Defence College of Management Technology,
UK Defence Academy, Cranfield University, United Kingdom.
Stuart H. Starr is President of the Barcroft Research Institute (BRI). Prior to
founding BRI, he was Director of Plans, MITRE; Assistant Vice President, C2 and
Systems Assessment, M/A-COM Government Systems; Director, Long Range Plan-
ning and Systems Evaluation, OASD(C3I), OSD; and Senior Project Leader, Institute
for Defense Analyses. He received his PhD in Electrical Engineering from the Univer-
sity of Illinois and was a Fellow at MIT’s Seminar XXI. Dr. Starr is a Fellow, Military
Operations Research Society (MORS); Associate Fellow, AIAA; Member of the Army
Science Board; a Senior Research Fellow at the Center for Technology and National
Security Policy, National Defense University; and a frequent participant in Blue Rib-
bon Panels of NATO, the National Research Council, and the Air Force Science Advi-
sory Board.
David Sulek is a Principal with Booz Allen Hamilton’s U.S. Security Team with
16 years of strategy, policy analysis, and general management consulting experience,
and has worked with the President’s National Security Telecommunications Advisory
Committee. He leads a team of policy analysts focused on homeland security, national
xvii
preparedness, information sharing, cyber security, and public-private partnership issues.

He received a master’s degree in national security studies from the Edmund A. Walsh
School of Foreign Affairs at Georgetown University, and a bachelor’s degree in politi-
cal science from Syracuse University.
Roelof Temmingh has been in the computer security industry for 15 years. In
2000, he co-founded SensePost as Technical Director and later was in charge of Re-
search and Development. At SensePost, he developed many successful security as-
sessment tools, including Wikto and Suru, and contributed to several books (Aggressive
Network Self-Defense, How to Own a Continent, Nessus Network Auditing). He is a
regular speaker at many of the top international security conferences (Blackhat, Defcon,
FIRST, CansecWest, RSA, etc.). At the start of 2007, he left the company to start Pa-
terva.
Olivier Thonnard is a PhD student at EURECOM (France), researching honeypot
traffic analysis under the supervision of Marc Dacier (Symantec Research). As a mili-
tary officer, he teaches at the Royal Military Academy, Polytechnic Faculty, Brussels.
His research interests include intrusion detection and network traffic analysis, specifi-
cally relating to the global analysis of Internet threats. Olivier is developing data min-
ing techniques based on clustering, correlation methods, and clique algorithms.
Tillmann Werner is a computer scientist at the University of Bonn. Previously, he
worked as an incident handler at the German national CERT. He is a member of the
Honeynet Project and has been doing research in the area of network-based attacks for
more than five years.
xix
Contents
Preface v
Kenneth Geers
About the Contributors xiii
Part I. Strategic Viewpoints
Cyber Wars: A Paradigm Shift from Means to Ends 3

Amit Sharma
Towards an Evolving Theory of Cyberpower 18
Stuart H. Starr
Sub Rosa Cyber War 53
Martin C. Libicki
Warfare and the Continuum of Cyber Risks: A Policy Perspective 66
Andrew Cutts
Cyber Terrorism: A New Dimension in Battlespace 77
J.P.I.A.G. Charvat
Borders in Cyberspace: Can Sovereignty Adapt to the Challenges of Cyber
Security? 88
Forrest Hare
Towards a Global Regime for Cyber Warfare 106
Rex Hughes
What Analogies Can Tell Us About the Future of Cybersecurity 118
David Sulek and Ned Moran
The Information Sphere Domain – Increasing Understanding and Cooperation 132
Patrick D. Allen and Dennis P. Gilbert Jr.
Sun Tzu Was a Hacker: An Examination of the Tactics and Operations from
a Real World Cyber Attack 143
Billy K. Rios
Belarus in the Context of European Cyber Security 156
Fyodor Pavlyuchenko
Politically Motivated Denial of Service Attacks 163
Jose Nazario
A Brief Examination of Media Coverage of Cyberattacks (2007 – Present) 182
Cyrus Farivar
xx
Part II. Technical Challenges and Solutions
Behavioral Analysis of Zombie Armies 191

Olivier Thonnard, Wim Mees and Marc Dacier
Proactive Botnet Countermeasures – An Offensive Approach 211
Felix Leder, Tillmann Werner and Peter Martini
When Not to Pull the Plug – The Need for Network Counter-Surveillance
Operations 226
Scott Knight and Sylvain Leblanc
Autonomic Computer Network Defence Using Risk State and Reinforcement
Learning 238
Luc Beaudoin, Nathalie Japkowicz and Stan Matwin
Enhancing Graph-Based Automated DoS Attack Response 249
Gabriel Klein, Marko Jahnke, Jens Tölle and Peter Martini
On nth Order Attacks 262
Daniel Bilar
Business and Social Evaluation of Denial of Service Attacks in View of Scaling
Economic Counter-Measures 282
Louis-Francois Pau
Virtual Plots, Real Revolution 294
Roelof Temmingh and Kenneth Geers
Subject Index 303

Author Index 305
Part I
Strategic Viewpoints
The Virtual Battlefield: Perspectives on Cyber Warfare 3
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-3
Cyber Wars:
A Paradigm Shift from Means to Ends
Amit SHARMAa,1
a
Institute for System Studies and Analysis (I.S.S.A),
Defence Research and Development Organization (D.R.D.O),
Ministry of Defence, India
Abstract. The last couple of decades have seen a colossal change in terms of the
influence that computers have on the battle field, to an extent that defence pundits
claim it to be a dawn of a new era in warfare. The use of computers and
information in defence has manifested into various force multipliers such as
Information Operations, C4I2SR Systems, Network Centric Warfare, to the extent
that commentators are terming this information age as a Revolution in Military
Affairs (RMA). These advances have not only revolutionized the way in which
wars are fought, but have also initiated a new battle for the control of a new
dimension in the current contemporary world: The Cyber Space.
Over time cyber warfare has assumed the shape of an elephant assessed by a group
of blind people, with every one drawing different meanings based upon their
perceptions. Under these circumstances there was a gradual paradigm shift in
military thinking and strategies, from the strategic aspect to the tactical aspect of
cyber warfare laying more emphasis on cyber attacks and counter measures. This
resulted in the formation of a notion that cyber warfare or information warfare is a
potent force multiplier, which in a sense downgraded the strategic aspects of cyber
war to a low grade tactical warfare used primarily for a force enhancement effect.
The author believes this is wrong, cyber war is a new form of warfare and, rather
than cyber war merely being an enhancement of traditional operations, traditional
operations will be force multipliers of cyber war.
This paper tries to shatter myths woven around cyber warfare so as to illuminate
the strategic aspects of this relatively misinterpreted notion. This paper will
elucidate the scenarios and mechanisms illuminating the process of using the
strategies of cyber war, so as to achieve conventional objectives. The paper will
also analyze the doctrine and strategies including first and second strike
capabilities with regard to cyber war. This paper identifies a paradigm shift from
the conventional belief of cyber warfare acting as a force multiplier for
conventional warfare to the recognition, that conventional warfare will be acting as
a force multiplier around cyber war and hence making cyber war as the primary
means of achieving grand strategic objectives in the contemporary world order.
Keywords: Cyber wars; cyber warfare; information warfare; strategy and doctrine.
The opinions expressed or implied in this paper are solely those of the author, and
1. are based on open sources. Under no circumstances these may be correlated or
perceived as the views of Government of India in general and the Author’s
organization in particular.
1
The Author is Deputy Director/Scientist ‘C’ in Information Security Division of Strategic Information
Dissemination Systems of I.S.S.A., D.R.D.O., Ministry of Defence, Government of India. The Author is a
Chevening Scholar from India and is currently pursuing M Sc. Global Security at UK Defence Academy.
Email: amitsharma.drdo@gmail.com and asharma.dcmt@defenceacadamy.mod.uk .
4 A. Sharma / Cyber Wars: A Paradigm Shift from Means to Ends
Introduction
“One hundred victories in one hundred battles is not the most skillful. Seizing the
enemy without fighting is the most skillful.”
Sun Tzu Sixth Century B.C [1]
Sun Tzu in sixth century B.C. eloquently referred to the fact that, the best form of
warfare is to take down the enemy without fighting with him.[2] Over time, as the
warfare has evolved, this notion has gained impetus, especially with the genesis of
cyberspace and cyber warfare. It was for the first time, that Sun Tzu’s notion of,
“Seizing the enemy without fighting is the most skillful” , could be imagined as
happening in its entirety, using this potent new weapon which, in current contemporary
world has no limits, no boundaries and to a surprise no visible restrictions or
legislations. Although over time the notion of information warfare has matured and
manifested into a form which has a colossal impact on how the contemporary wars are
fought, but this has also resulted in the downgrading of strategic side of information
warfare or cyber warfare to a decisive tactical force multiplier capable of turning the
tides in war. Whilst this force enhancer aspect of cyber warfare is an important and
decisive component of conventional warfare, but against the conventional wisdom this
is not the end, but merely a beginning of the strategic aspect of cyber warfare.
In order to analyze the strategic aspect of cyber warfare, Luttwak’s criteria of
integration of a strategic warfare across all spectrum of affairs right from the tactical to
the grand strategic level,[3] provides an important criterion for postulating the strategic
framework for cyber warfare; Or in terms of Liddell-Hart, the coordination and
assessment of means to achieve ends at all levels plays [4] a dominant role in casting a
cyber warfare strategy. In light of these considerations, the author will elucidate the
framework in which cyber warfare will have a strategic effect by acting as primary
means to achieve conventional ends, hence will induce a paradigm shift from the
conventional notion of cyber warfare as a tactical force multiplier to the notion of
strategic cyber warfare acting as primary means of achieving grand strategic objectives
in the contemporary world order. The author will accomplish this objective by deriving
the elixir of Clausewitz’s Trinitarian warfare and applying the concepts of Rapid
dominance and Parallel warfare in cyber space so as to generate the strategic paralytic
effect envisaged in effect based warfare. The author will conclude by shattering the
conventional dictum of cyber defence, based on the notion of “defence in layers” and
legal aspects of Law of Armed Conflict; by providing the only feasible and viable
cyber defence strategy relying on the application of Rational Deterrence Theory (RDT)
in general and on the idea of Mutually Assured Destruction (MAD) in particular so as
to maintain the strategic status quo.
A. Sharma / Cyber Wars: A Paradigm Shift from Means to Ends 5
1. Cyber Irony- The Revolution in Military Affairs
Over last couple of decades, Information assets have had an irrefutable impact on the
way, in which conventional wars are fought, to an extent that military theorists have
termed it to be a Revolution in Military Affairs (RMA).[5] This extensive reliance of
conventional warfare on information in contemporary conflicts is often misrepresented
as information warfare rather than information-enabled warfare or information-
enhanced warfare. The sudden significance ushered to this relatively new paradigm of
information-enabled warfare, where information warfare is acting as a decisive force
multiplier, has also raised certain existential questions for its survival predominantly
that; whether this new paradigm of Information enabled warfare is really a strategic
information warfare which will be the primary means of achieving ends or is it just a
misinterpreted notion created by its loyal supporters, only to pacify the appetite of the
change-hungry military world?
The answer to this existential question lies in the debate revolving around the
notion, that information warfare is a revolution in military affairs. Alvin and Heidi
Toffler have termed this information-enabled warfare as a third wave[6]; similarly most
of the contemporary military theorists have termed this misinterpreted information
warfare as a revolution in military affairs. At this juncture an important argument
looms around the relation of a revolution in military affairs and its strategic effect.
Throughout the history whenever there has been an occurrence of revolution in military
affairs, it has always been followed with a strategic effect; for example revolutions in
military affairs such as guns, artillery, airpower, nuclear weapons and so on, have
always been accompanied with their strategic impact in creating a new world order.
This important relationship between revolution in military affairs and its strategic effect
is clearly missing in case of information warfare.
Hence if information warfare is really a revolution in military affairs then ideally it

should have a strategic effect and since that effect is clearly missing, it can be
concluded that something somewhere is missing. This gap is due to the
misinterpretation of information warfare as mere decisive tactical information-enabled
warfare acting as a force multiplier for conventional warfare. The Author believes that
this notion is a fallacy; information warfare is more than just information-enabled
warfare, which albeit represents an important aspect of information or cyber warfare,
but not in totality. Cyber warfare is a strategic warfare which can be used as a principle
means to achieve strategic ends and as required by Luttwak’s criterion for strategic
warfare [7], the framework for the strategic cyber warfare is to be defined across all
spectrum of affairs right from the grand strategic to the tactical level.
2. The Grand strategic cyber warfare – the triad theory of cyber warfare
“War is thus an act of force to compel our enemy to do our will”

Clausewitz [8]
Clausewitz in his book On War clearly elucidated the fact that the end of the war is to
compel the enemy to do your will [9] and Sun Tzu argues that the best form of warfare
is the one in which the enemy is seized without a fight [10].Cyber warfare derives the
essence of both of these great military theorists as it is a warfare which is capable of
compelling the enemy to your will by inducing strategic paralysis to achieve desired
ends and this seizing of enemy is done almost without any application of physical force.
Clausewitz formulated the theory of nature of war based upon the conception of
Trinity. This elusive Trinitarian warfare according to Clausewitz held the key to victory
in a war. Clausewitz predominantly constructed this trinity around three dominant
tendencies, the blind force composed of primordial violence, hatred and enmity; the
play of uncertainty and chance in which the creative spirit roams; and the reason for
violence or the political instrument.[11] The tendencies are abstracted as, the people or
the will to fight a war in terms of finances, manpower and support; the military or the
means; and the government or the effort, the leadership and the direction which is
essential for a nation.
These three tendencies extensively interact with each other and have a
continuously changing relationship. Till the time they are present and are interacting,
the nation will sustain even when hit with the worse case scenarios. It is only when all
of the three components are destroyed together or in conventional terms are subjected
to parallel warfare; it is only then a ‘cascade effect’ will be generated to induce a
strategic paralytic effect onto the Nation and the Nation as a system will crumble
resulting in chaos and mayhem. In the current contemporary world in general and the
developed countries in particular, the reliance on modern technology is not treated as a
luxury, but a necessity where all the three tendencies are extensively dependent upon
cyber space in one form or the other (Figure 1).
The modern militaries extensively depend on information assets and cyber space
especially in scenarios where the deployment is across the globe. These information
assets are used extensively in command and control systems especially in joint or
coalition operations; in net-centric warfare operations involving global information
grids; for logistic; for surveillance right from the information gathering which requires
data links with satellites to information dissemination involving strategic information
dissemination system; for communication right from the tactical field or theater data
link operations and networks to strategic command and control networks involving
communication satellites; global positioning and navigation satellites and networks for
not only navigation, but for precision targeting; and so on(Figure 1).
The scenario is the same with the tendency involving the people or the will to
fight a war in terms of finances, manpower and support. In almost all the developed
countries and in some developing countries, people rely extensively on computers and
cyber assets for almost all of their daily chores. Whether utilities such as gas and
electricity; or health, transportation and banking facilities, all rely on cyber or network
assets for their functioning. The scenario is even worse in case of banking and
economic institutions like stock exchanges where money, which is nothing but
numbers or information, travels across the national boundaries to the remotest corner of
the world due to globalization of financial and banking instruments of economy. These
technological advantages have not only eased the life of people, but have also made
them vulnerable; as day by day people are becoming hopelessly dependent on these
facilities to an extent that they take them for granted. The media and communication
networks have almost become the vital sensory organs of this technologically
dependent society.
This western society according to Bill Durodie is becoming more and more
individualistic. It is becoming a society where people are socially disconnected;
politically disengaged; and are in scientific disbelief.[12] In this society where
perceptions overweight the reality, the people are becoming more and more ‘risk
averse’ and are constantly living in an environment of fear. This state of society is
classically defined by Ulrich Beck as a Risk Society.[13]
These societies are socially disconnected; politically disengaged; in scientific

disbelief; and are constantly living in an environment of fear. The sudden
disappearance of almost all of their facilities on which they are hopelessly dependent
upon, will result in catastrophic outcomes where chaos, fear, bedlam, anarchy and basic
animal instincts of man will prevail resulting in mayhem and complete destruction of
nation as a system (Figure 2).
Figure 1: The notion of trinity in terms of strategic cyber warfare. (Source: Author)
The third and the most dominant tendency constitute the government, the political
instrument and the leadership. Government plays an important role in providing
leadership to the nation and in combining efforts and means to achieve political aims;
hence for the failure of nation as a system, it is important that along with the other two
tendencies the political instrument should also be destroyed.
In current contemporary world, governments play as the political instrument in the

trinity by means of excising control and gaining the mandate of people. These
objectives are achieved by using effective law enforcement and by providing a secure,
secular and democratic environment to people to attain control and mandate over them.
The law enforcement and security agencies rely extensively on criminal records and
other coordination networks such as emergency response and recovery networks which
although act as a force multiplier for them, but at the same time make them vulnerable
to strategic cyber warfare. Another important aspect to gain mandate and control of
people is the media. Media is an important tool that frames the perception and
psychological frame of mind of the population. The ‘CNN effect’ is a potent tool to
influence the mindsets of people.[14] Media around the world is extensively interlinked
through networks which not only makes information to disseminate easily, but also
make media more susceptible to strategic cyber warfare. These Media networks can be
hijacked for tarnishing the image of politicians and the government of a victim nation;
and can be used to induce fear and chaos among people. PSYOPS can be fully
employed on these hijacked media networks to tarnish the political stance of the victim
nation at national and international levels, thus initiating conflicts both at inter-nation
level (by routing an attack through a victim nation) and intra-nation level (by inducing
political divisions in the population by false propaganda and so on, thus resulting in
conflicts and total law and order failure), hence creating the symptoms of a failed state
which has anarchy, fear and chaos, resulting in a humanitarian crisis and failure of the
state (Figure 2).
As described above, these three tendencies form the core of a nation or constitute
the nation as a system of systems. All of these components are quite resilient in nature
unless and until they are simultaneously attacked and destroyed, they are quite capable
of reviving one another. Thus in order to achieve a strategic effect and to gain rapid
dominance, parallel warfare should be initiated in cyber space so as to destroy all the
three tendencies simultaneously (Figure 2). The Author believes that the cyber attacks
such as Titan Rain; attacks on Estonia and Georgia; and so on were not successful due
to the fact that they were tactical in nature and were targeting the individual
components of the trinity at a time. Since these components are resilient when they are
together, so even if one of the tendencies is fully destroyed in a cyber attack, the other
two tendencies will tend to rescue it, hence a strategic paralytic effect will not be
achieved. The author believes that it is for this reason the cyber attacks conducted so
far could not achieve a strategic effect and were ineffective.
Figure 2: Cyber Trinity based parallel cyber warfare attack to induce strategic paralytic effect on a victim
nation (Source: Author)
In order to achieve a strategic paralytic effect via the application of cyber warfare,
all the three components of the trinity should be attacked simultaneously. These cyber
attacks should be performed by using the paradigm of parallel warfare[15], which relies
on gaining rapid dominance by producing the desired effect of paralyzing the control of
the enemy by performing rapid decisive operations at all levels i.e. the strategic,
operational and tactical; across the spectrum involving related assets and critical
components of all the three tendencies of the cyber trinity; and with rapid succession so
as to reduce the chances of counter attack or of the “defensive phenomena” of pulling
the plug.
When the art of simultaneous parallel warfare to achieve rapid dominance is

combined with the strategy to simultaneously attack the three components of the cyber
trinity, it will generate a cascade effect rendering the victim in a paralytic state with the
loss of control over the state and failure of the state-as-a-system, thus generating
ramifications which are way ahead than mere arithmetic benefits generated by a
successful attack. This failure of Nation as a system-of-systems and the disruption or
paralysis of the state will destroy the victim’s will and capability to fight thus
‘compelling it to submit to your will’.
This strategy of strategic cyber warfare against the Trinity in cyber space to
achieve strategic paralysis provides for an alternative warfare or means to achieve
strategic effect of rendering the enemy ineffective to operate as per its wishes,
eventually is more important than the conventional paradigm of destruction-based
warfare to annihilate the forces it depends upon for its defence, hence generating not
only a strategic victory, but also a constructive conflict termination. This constructive
conflict termination is not only desired, but is imperative especially in the
contemporary world, with examples of conflicts with flawed exit strategies resulting in
victories turning into protracted wars such as the Iraq war.
3. Campaign planning- The orchestration of strategy
Once the strategy for conducting cyber warfare to achieve the strategic end of
compelling the enemy to submit to your will by rendering the enemy ineffective is
defined, the next important task is to integrate that strategy to a campaign plan which
spreads across all levels of warfare. This process of accessing the assets and means to
achieve the desired effect and ends by orchestrating the strategy across all levels of
warfare is termed as the campaign planning and the scenario is the same in pursuing a
strategic cyber warfare campaign.
As done in conventional warfare the campaign planning for strategic cyber warfare
is also based on phasing, but against the conventional dictum of executing these phases
in sequential or near sequential manner, the strategic cyber warfare tends to use these
phases in almost near parallel manner. At any temporal instance, each of the phases
will have a substantial effect over the entire theater of operations in general and
individual zones in particular in conformance with the parallel warfare dictum. When
applied in order to orchestrate the strategic cyber warfare strategy of generating a
paralytic effect on the adversary by initiating parallel cyber attacks on all the three
components of the Trinity, all phases of the campaign will overlap extensively and the
campaign will be in the form of simultaneous waves of these overlapping phases
tailored according to the theater conditions at all levels.
The campaign for strategic cyber warfare in line with conventional dictum[16] will
consist of five broad stages; Shape, Deter, Seize initiative, Dominate and Exit (Figure
3). Out of these the Shape and Deter are part of the pre conflict phases; Seize initiative
and Dominate are usually the conflict phases; and the Exit is the post conflict phase.
Although these phases are lucidly categorized as pre-conflict; conflict; and post-
conflict phases, these categorizations tend to overlap extensively. For example even
though there would be a conflict in progress in certain parts of the theater, but still in
certain other parts, the Shape and Deter phases may be exercised to limit the conflict
from further escalation.
The initial phase or the Shape phase revolves around shaping the conflict.[17] This
is done using extensive peacetime cyber reconnaissance, such as mapping enemy’s or
potential adversary’s cyber assets, network design, layout, vulnerabilities, critical
components and dependence; assessing enemy’s cyber defence capabilities both
offensive and defensive; boosting national cyber defence capabilities not only in
military, but in all the components of the cyber trinity; identification of critical
assets/targets which would initiate a ‘cascade effect’. In order to identify these
components the ‘Critical component theory’ which was utilized by the allied air force
for strategic bombing of Germany during the Second World War[18] can prove an
important asset, And performing all the above operations on your own assets also, can
be useful to identify potential susceptibility and vulnerabilities.
Apart from cyber reconnaissance, an important component included in the

peacetime operations involves the process of inducing vulnerabilities in enemy’s cyber
assets. This is achieved either by using cyber means such as installing covert malware
such as trojans, rootkits, dormant stealth malware and so on; or by using covert means
such as exporting bugged firmware by using front door companies, thus making enemy
systems susceptible to Permanent Denial Of Service (PDOS) attacks and by covertly
gaining information of enemy’s critical system software such as operating systems, by
gaining access to the skeleton keys for the backdoors, usually channeled through
vendor influence; and so on.
Deter like the Shape phase is also a pre-conflict phase which extensively revolves
around shaping the future conflict by gaining a credible and known deterrence. This
phase capitalizes on the information gained during the cyber reconnaissance of the
adversary’s and of personal cyber infrastructure and their susceptibility to strategic
cyber attack. Based upon this information, relevant steps are initiated to harden
personal infrastructure in terms of ‘layers of defence’ and redundancy; and to prepare
for exploiting the enemy’s vulnerabilities and then testing them in simulated
environments by conducting cyber war games.
An important part of this phase is to develop a cyber deterrence which is credible

and is made known to the enemy. The credibility of the cyber deterrence can be
achieved by creating a Cyber Triad capability, equivalent to a Nuclear Triad[19] which
will have capability for orchestrating a second strike in case of failure of the deterrence.
Cyber Triad capability can consist of Regular defence/military assets and networks as
forming the first section of the triad; the second section of the triad can consist of an
isolated conglomerate of air-gapped networks situated across the friendly nations as
part of cooperative defence, which can be initiated as credible second strike option; and
the third section of the triad can consist of a loosely connected network of cyber militia
involving patriotic hackers, commercial white hats and private contractors which can
be initiated after the initial strike or in case of early warning of a potential strike. This
Cyber Triad creates a scenario of a credible and undisputable cyber deterrence and
second strike capability thus assuring a Mutually Assured Destruction (MAD) in cyber
space.
The later feature of the cyber deterrence involving the policy of making the cyber
deterrence known to the enemy, can be achieved by following a ‘cyber countervailing’
strategy in line with the ‘countervailing’ nuclear strategy followed by NATO forces
during the cold war.[20] This strategy revolved around making known to the potential
adversary that the implication of a nuclear strike would be far greater than the potential
gains an adversary can achieve by initiating the first strike. The scenario will remain
the same in the strategic cyber warfare where the potential enemy should be made
known of the potential risk it might be facing in light of initiating a first strike. This can
be achieved by means of media coverage; extensive war games; and to some extent by
covert instantiation of limited cyber warfare attacks on the adversary. The author
believes that the recent attacks such as the Titan Rain; the attacks on Estonia and
Georgia; and attacks on various other countries around the world such as UK, France
Germany, India and so on; can be a part of a cyber countervailing strategy.
Once the conflict has been initiated there can be two possibilities; either the cyber
deterrence has failed and the enemy has initiated the first strike or due to certain
unforeseen events, friendly forces needed to initiate the first strike. In the former
situation it should be always assumed that the adversary’s first strike will have, if not
an all out decapitating effect, a certain degree of effect on friendly cyber infrastructure,
but an important aspect here is to detect the attack before it could generate any strategic
paralytic effect. Once this detection is achieved, it is required that necessary steps
should be taken so as to activate the cyber triad to initiate the second strike capability.
This can be achieved by taking defensive counter measures and securing cyber
infrastructure; initiating the second triad involving a coordinated and real-time
integration of an isolated conglomerate of air gapped networks situated across friendly
nations as part of cooperative defence; and in a worst case scenario, where there is a
total loss of offensive and defensive capabilities, the third component of the triad
consisting of a loosely connected network of cyber militia involving patriotic hackers;
commercial white hats and private contractors should be activated in order to initiate a
Figure 3: Cyber campaign planning for strategic cyber warfare. (Source Author)
protracted conflict. This protracted conflict is essential to gain time so as to revitalize

your own capabilities in worst case scenarios.
Similarly if the first strike is to be initiated, the primary aim should be on the total
destruction of the cyber triad. This is done so as to destroy enemy’s capability of
retaliation and to provide strategic freedom of operation in cyber space to friendly
forces. This strategic freedom of operation in cyber space is essential in order to initiate
the strategic cyber warfare on the critical components of the cyber trinity for achieving
the strategic paralytic effect. An important consideration at this point should be on the
fact that these phases should be conducted in an overlapping fashion to perform parallel
warfare in cyber space.
Once the necessary objectives are achieved and the desired end-state is reached
due care should be taken to stabilize the situation. Although the effects of strategic
cyber warfare like the nuclear warfare are far reaching, devastating, and in a way
cannot be measured, they still can be contained and the post-conflict situation can be
stabilized by initiating external support in the form of reestablishing the cyber services
and facilities and retrieving of data, if not fully, then to a decent level, so as to carve
out an exit strategy. Like in conventional conflicts, the chances of a cyber insurgency
consisting of patriotic hackers and of humanitarian crisis loom at large especially in a
post-conflict scenario. Hence due care should be taken to reestablish the essential
services and command and control infrastructure after the conflict. An aggressive
media strategy to counter dissident feeling and anguish among the people should be
utilized so as ease the stabilization process and assist with post conflict rehabilitation.
4. Cyber Defence – A conventional fallacy
The current conventional wisdom on cyber defence relies on the notion of ‘defence in
layers’[21] and on International legal regulations especially by drawing similarities
between cyber attacks and armed conflicts and then applying the law of armed
conflict [22] appropriately. The notion of ‘defence in layers’ is a tried and tested
dictum which is extensively used to protect both the commercial and the defence
networks. It relies on installing multiple layers of defences so as to make the
penetration almost impossible. Even though this notion has extensively been used to
protect cyber infrastructure, it is a known fact that such a system is as strong as its
weakest link. No matter how much the system is hardened and no matter how many
layers are used to secure the system, there is still no guarantee that the system security
is foolproof. It is safe only up until the time when someone doesn’t find any
vulnerability or an exploitable construct in the system, which can be exploited to gain
access in to the system. Yes, this notion of defence at least assures that the penetrator
will require time to defeat multiple layers of security. It is this time that is crucial for
defenders to take necessary action to thwart the threat. Hence this provides for a
minimum deterrence, but nevertheless is not a complete and foolproof solution.
The other aspect of conventional wisdom on cyber defence relies on the legal
framework of international law both jus ad bellum and jus in bello. This defensive
strategy relies on the deterrence generated by the legal punitive aftermath of a cyber
conflict on the erring side. The notion relies extensively on drawing similarities
between the conventional international armed conflicts and cyber conflicts; and then
applying the international laws which will fit the scenario. Most of the discussion in
this realm revolves around self defence for jus ad bellum and Law of armed conflict for
jus in bello. Extensive analysis is done in order to find the lines of similarity between
the armed conflicts and cyber attacks, but these arguments in a sense degrade the
strategic aspect of cyber warfare to mere tactical cyber attacks or are centered towards
the means rather than the strategic ends.
The author believes that this notion of defence is a fallacy as the underlying
assumption of treating strategic cyber warfare as mere tactical cyber attacks is in itself
a fallacy. If cyber warfare is performed so as to achieve the strategic paralytic effect,
then the consequences of such a warfare would be far reaching and to an extent not
measurable in conventional terms. The cascade effect initiated as an aftermath of
strategic cyber warfare would generate a chain of ‘unintended consequences’ that are
almost impossible to tackle using the conventional framework of law of armed conflict.
The only appropriate legal frame work to handle strategic cyber warfare would be
based on the legal frame work of jus ad bellum and jus in bello for nuclear weapons,
which unfortunately is a long debated notion and has an incoherent international
opinion visible extensively in the ICJ’s opinion [23] over the use of nuclear weapons.
In light of these circumstances, where the conventional dictum of cyber defence is

a mere fallacy the only viable and achievable option for cyber defence would rely on
the age old dictum of deterrence and to an extent on the cold war principle of Mutually
Assured Destruction. This cyber deterrence can be guided by the Rational Deterrence
Theory (RTD) which relies on the underlying assumption of actors to be rational and
performing cost benefit analysis before reaching any logical conclusions; and the
outcome variation depends upon the variations of opportunities which the antagonists
have.[24] Ashen and Snidal argue that the key concepts for achieving deterrence based
on RDT will be on the credibility of the deterrence capabilities and the rational actor
assumption of decisions relying on cost benefit analysis.[25] In terms of cyber
deterrence the credibility can be achieved by the creation of a cyber triad as part of the
Deter phase of cyber campaign planning. This cyber triad capability can consist of
Regular defence/military assets and networks as forming the first section of the triad;
the second section of the triad can consist of an isolated conglomerate of air-gapped
networks situated across the friendly nations as part of cooperative defence; and the
third section of the triad can consist of a loosely connected network of cyber militia
involving patriotic hackers, commercial white hats and private contractors
This credible second strike capability assures the dictum of Mutually Assured
Destruction (MAD) in cyber space and hence an option for defence in terms of
deterrence. This capability should be made known to the potential advisories as part of
cyber countervailing strategy to warn them of undesired consequences and punitive
costs they may bear in the event of a cyber conflict. This form of deterrence is
generally classified as the deterrence by punishment; the other form of deterrence is
classified as the deterrence by denial.[26] This deterrence by denial in cyber defence
can be achieved by preemptive cyber strikes on the adversary’s cyber offensive
capabilities. However, in scenarios of state actors this policy may result in a further
escalation of conflict; hence utmost care and thought process in regard to attribution of
cyber attacks should be taken before initiating such a strike. In the case of non-state
actors the deterrence by denial in the form of preemptive cyber strikes, offer a credible
deterrence mechanism for thwarting any such threat. In both the scenarios, extensive
reconnaissance and surveillance both by cyber and conventional means can act as
suitable tools for attribution and target selection.
It follows that cyber deterrence can act as an important means for thwarting both
the state and non-state threats by means of deterrence by punishment and deterrence by
denial. Also it is clearly evident that none of the cyber defence notions can provide for
a holistic cyber defence; hence a cyber defence strategy should be a combination of the
notion of ‘defence in layers’; the legal aspects of International law, although whether
the Law of Armed conflict may be applicable in its entirety is debatable, but that is out
of the scope for this article; and by generating a credible cyber deterrence based on the
cyber triad, thus assuring a Mutually Assured Destruction in cyber space and hence a
strategic status quo.
5. Cyber finale
The last couple of decades have seen a colossal change in the way in which
conventional wars are being fought, with information being an integral component.
Information and information assets have made such an indelible impact on warfare that
military pundits often misinterpret this information-enabled warfare to be information
warfare, where information assets act as decisive force multipliers, which are capable
of changing the outcome of wars. Although it is a considerable feat, but in a sense it
degrades the strategic aspect of information warfare to mere tactical cyber attacks.
Over years this paradigm of considering information warfare as mere tactical force
multiplier has gained impetus. This paradigm has created an environment where
“means are emphasized more than the desired ends”. The author believes that this is a
fallacy and calls for a paradigm shift from “means to ends”. The Information warfare is
a strategic warfare which derives the essence of both Sun Tzu and Clausewitz as it is a
type of warfare which is capable of compelling the enemy to do your will by inducing
strategic paralysis to achieve desired ends and this seizing of the enemy is done with
virtually no application of physical force.
The strategic information warfare is capable of achieving desired strategic ends by

inducing a strategic paralytic effect onto the nation; and the nation as a system will
crumble resulting in chaos and mayhem. This strategic effect relies on the framework
defined across all spectrum of affairs, right from the grand strategic to the tactical level,
and is achieved by the strategy of strategic cyber warfare against the Clausewitz’s
Trinity in cyber space to achieve strategic paralysis and rapid dominance using parallel
warfare. This strategy provides alternative means to achieve the strategic effect of
rendering the enemy ineffective to operate as per its wishes, which eventually is more
important than the conventional paradigm of destruction-based warfare to annihilate the
forces, the enemy depends upon for its defence; hence generating not only a strategic
victory, but also a constructive conflict termination.
In order to achieve these ends, a comprehensive orchestration of strategy in the

form of campaign planning is required, which in line with conventional dictum will
consist of five broad stages; Shape, Deter, Seize initiative, Dominate and Exit;
categorized further as pre-conflict, conflict and post-conflict phases. Out of these the
most crucial are the pre-conflict phases of Shape and Deter, which involve extensive
cyber reconnaissance and the creation of a credible and known cyber deterrence based
on cyber triad capability and cyber countervailing strategy. This creates a credible
second strike option in cyber space, which assures a strategic status quo based on
Mutually Assured Destruction in event of a cyber conflict.
In terms of cyber defence the conventional wisdom of treating of cyber warfare as

mere tactical cyber attacks or force multipliers and relying on the legal framework of
Law of Armed Conflict (LOAC) for defence is a fallacy; as it not only undermines the
strategic aspect of cyber warfare in the form of generating strategic paralytic effect to
achieve political ends, but it also relies on a legal framework for defence, which is
contemplated on a false assumption of drawing similarities with conventional armed
conflicts. The only warfare which matches cyber warfare in strategic terms is nuclear
warfare; and as there is a gross division of International Law in terms of Jus ad bellum
and jus in Bello for Nuclear weapons, the scenario unfortunately will remain the same
for cyber warfare also.
Under these circumstances, the only feasible and viable cyber defence strategy will
rely on the application of Rational Deterrence Theory in general and on the idea of
Mutually Assured Destruction in particular so as to maintain the strategic status quo.
Hence the author recommends that the nations should reconsider their cyber defence
strategies, and should define a strategy based on a combination of the notion of
“defence in layers”; legal instrument; and potent cyber triad based cyber deterrence.
This would be the only viable and achievable option in current contemporary and
futuristic conflicts, which the author predicts, will be dominated by strategic cyber
warfare as the “primary means to achieve strategic ends”.
References
[1] Sun Tzu translated by Samuel B. Griffith, The Art of War. (Oxford University Press, 1963).
[2] Sun Tzu translated by Samuel B. Griffith, The Art of War. (Oxford University Press, 1963).
[3] Luttwak Edward, Strategy: The Logic of War and Peace, (London: Harvard university press, 1987).
[4] Liddell Hart Basil Henry, Strategy, (New York: Meridian, 1991).
[5] Charles A. Ray, “Cyber war and Information Warfare: A Revolution in Military Affairs or Much Ado
about Not Too Much?”, National War College Report, 1997.
[6] Alvin and Heidi Toffler, The Third Wave- The Classic Study Of Tomorrow, (Bantam Books,1980)
[7] Luttwak Edward, Strategy: The Logic of War and Peace, (London: Harvard university press, 1987).
[8] Michael Howard and Peter Paret, eds., On War (Princeton University Press, 1984) Book I, Chapter 1,
section 2, pp 75.
[9] Michael Howard and Peter Paret, eds., On War (Princeton University Press, 1984) Book I, Chapter 1,
section 2, pp 75.
[10] Sun Tzu translated by Samuel B. Griffith, The Art of War. (Oxford University Press, 1963)
[11] Michael Howard and Peter Paret, eds., On War (Princeton University Press, 1989) pp 583.
[12] Bill Durodie, “The Limitation of Risk Management” , TIDSSKRIFTET POLITIK ,Vol. 8 No.1,2004.
[13] Ulrich Beck, Risk Society: Towards a New Modernity, (New Delhi: Sage, 1992).
[14] Belknap, Margaret H. The CNN Effect: Strategic Enabler or Operational Risk?, U.S. Army War
College Strategy Research Project. 2001.
[15] Deptula David, Firing for Effect: Change in the Nature of warfare,(Arlington: Aerospace Education
Foundation,1995).
[16] Joint Publication 3-0 Doctrine for Joint Operation, December 2005,IV-34.
[17] Joint Publication 3-0 Doctrine for Joint Operation, December 2005,IV-34.
[18] Pape Robert, Bombing to Win: Airpower and coercion in war,(Ithaca: Cornell university press,1996).
[19] Aldridge Robert C, First strike!: the Pentagon's strategy for nuclear war,(Cambridge: South End
press,1983)
[20] Powell Robert, ”Nuclear Deterrence and the Strategy of Limited Retaliation”, The American Political
Science Review, Vol. 83, No. 2 (Jun., 1989), pp. 503-519
[21] Harold F. Tipton and Micki Krause, Information Security Management Handbook, (Florida: CRC Press,
2006).
[22] Matthew E. Haber, Computer Network Attack and the Laws of Armed Conflict: Searching for Moral
Beacons in Twenty-First-Century Cyber warfare, ARMY Command and general staff college, Fort
Leavenworth, 2002.
[23] The 1996 ICJ’s opinion over the Legality of Nuclear Weapons.
[24] Achen, Christopher and Snidal Ducan,” Rational Deterrence Theory and comparative case studies”,
World Politics, Vol. XLI, No. 2, pp.139-169.
[25] Achen, Christopher and Snidal Ducan,” Rational Deterrence Theory and comparative case studies”,
World Politics, 1998, Vol. XLI, No. 2, pp.139-169.
[26] Downs George,” The Rational Deterrence Debate”, World Politics, 1998, Vol. XLI, No. 2, pp.225-237.
18 The Virtual Battlefield: Perspectives on Cyber Warfare
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-18
Towards an Evolving Theory of

Cyberpower
Dr. Stuart H. STARR1,a
a
Center for Technology and National Security Policy (CTNSP)
National Defense University (NDU)
Abstract. In the 2006 Quadrennial Defense Review, a request was made to have
the Center for Technology and National Security Policy (CTNSP), National
Defense University (NDU), develop a theory of cyberpower. It was noted that
there was a need to develop a holistic framework that would enable policy makers
to address cyber issues in proper perspective.
To satisfy that tasking, CTNSP convened five workshops, drawing on experts from
government, industry, academia, and think tanks. Those workshops addressed a
broad set of issues related to the evolution of cyberspace, cyberpower,
cyberstrategy, and institutional factors that influence those factors (e.g.,
governance, legal issues).
To develop the desired theory, this paper systematically addresses five key areas.
First, the paper defines the key terms that are associated with cyber issues.
Particular emphasis is placed on the terms “cyberspace”, “cyberpower”, and
“cyberstrategy”. Second, the paper categorizes the elements, constituent parts, and
factors that yield a framework for thinking about cyberpower. Third, the paper
explains the major factors that are driving the evolution of cyberspace and
cyberpower. To support that effort, the paper presents strawman principles that
characterize major trends. Fourth, the paper connects the various elements of
cyberstrategy so that a policy maker can place issues in proper context. Finally, the
theory anticipates key changes in cyberspace that are likely to affect decision
making.
In view of the dramatic changes that are taking place in cyberspace, it is important
to stress that this effort must be regarded as a preliminary effort. It is expected that
the theory will continue to evolve as key technical, social, and informational trends
begin to stabilize.
Keywords: cyberspace, cyberpower, cyberstrategy, cyber institutional factors
Introduction
This white paper represents a continuing effort to evolve a theory of cyberpower. The
white paper begins by characterizing the components of a “theory of cyberpower”.
Consistent with that characterization, we identify key terms and put forth straw man
definitions of those terms. We then identify the specific objectives that will be
1
Disclaimer: The views expressed in this article are those of the author and do not reflect the official
policy or position of the National Defense University, the Department of Defense or the U.S. Government.
All information and sources for this paper were drawn from unclassified materials.
S.H. Starr / Towards an Evolving Theory of Cyberpower 19
addressed in this theory. In accord with those objectives, we present a holistic

framework to categorize and discuss key categories. Within this holistic framework, we
discuss the intellectual capital required to address these issues.
Subsequently, we discuss theoretical dimensions of the key categories: cyberspace,
cyberpower, cyberstrategy, and institutional factors. In addition, we discuss the
challenges associated with connecting across these categories and anticipating the
future cyber activities and issues of interest.
We conclude the white paper by summarizing major findings and identifying the
next steps that should be taken to refine this evolving theory of cyberpower.
1. Context
To provide context for this white paper, this section discusses elements of a theory,
objectives, approach, structure, key definitions, and required intellectual capital.
1.1. Elements of a Theory
A theory of warfare should address five key issues [1]. First, it should introduce and
define the key terms that provide the foundation of the theory. Second, it should give
structure to the discussion by categorizing the key elements of the theory. Third, it
should explain the elements in these categories by summarizing relevant events and
introducing key frameworks or models. Fourth, it should connect the various elements
of the subject so that key issues can be treated comprehensively. Finally, it should seek
to anticipate key trends and activities so that policy can be germane and useful.
This framework for a theory raises one immediate issue. There is interest in the
ability to predict, rather than anticipate, key activities. However, as described below,
the cyber problem is in the midst of explosive, exponential change. In the midst of this
exceptional uncertainty, it is infeasible to make reliable predictions. Thus, we have
adopted the less challenging task of “anticipating” key trends and activities.
Finally, it is important to stress the following caveat: since this is an evolving
effort to develop a theory of cyberpower, the emerging theory will not be complete.
Furthermore, as discussed below, early efforts to develop a theory for a discipline have
inevitably been somewhat wrong.
To provide some context for theoretical developments, it is useful to note the
challenges that the theories associated with physics have faced in its evolution.
Contemporary physics theory has evolved over hundreds of years, dating back to the
seminal contributions of Galileo and Newton. In this discipline, there is a common base
of knowledge, although there are significant variants for specific sub-areas (e.g.,
quantum mechanics, classical dynamics, relativity). In addition, there are strong links
to other “hard science” disciplines (e.g., math, chemistry, biology). Although the
definitions of key terms and concepts are generally established, it should be noted that
there were many false starts (e.g., a hundred years ago, physicists had (incorrectly)
postulated the existence of an ether through which electromagnetic waves propagated
as they traversed a vacuum). Even in contemporary times, discussions still persist about
the fundamental definitions of matter (e.g., quarks with a variety of properties).
Within the sub-areas of physics, there is broad agreement about key categories
(e.g., solid, liquid, and plasma physics). In these key sub-areas, mathematical models
20 S.H. Starr / Towards an Evolving Theory of Cyberpower
have generally been developed drawing on experiments and observations. Many of

these mathematical models have proven to be extremely accurate and precise in
explaining and predicting outcomes. However, there are still efforts underway to
connect many of the key sub-areas of physics. For example, there is considerable work
underway in the area of “string theory” to develop a unified understanding of basic
phenomena, although some critics have argued that this is likely to be a dead end [2].
To highlight the challenges facing the “cyber theorist”, it is useful to contrast the
discipline of physics with that of cyberspace. The cyberspace of today has its roots
back in the 1970s when the Internet was conceived by engineers sponsored by ARPA.
Detailed analysis of cyberspace issues often requires even broader cross-disciplinary
knowledge and skills than physics. These include, inter alia, computer scientists,
military theorists, economists, and lawyers. Each of these disciplines has its own
vocabulary and body of knowledge. Thus, it is quite challenging for these stakeholders
to communicate effectively. This is manifested in debates about the most basic of terms
(e.g., “cyberspace”) where key definitions are still contentious. Consistent with the
heterogeneous nature of the problem, it is not surprising that prior efforts to
characterize this space have not been successful. At present, there is no agreed upon
taxonomy to support a comprehensive theory.
As noted above, key attributes of a theory include its ability to explain and predict
(or at least, to anticipate). There are many reasons why prior theoretical cyber efforts
have foundered. These include the facts that key facets of the field are changing
exponentially, there is little or no agreement on key frameworks, and the social science
element of the discipline (e.g., understanding of cognition, human interactions in
virtual societies) makes it very difficult to develop models that reliably explain or
anticipate outcomes. Finally, we are unable to connect the disparate elements of the
field because a holistic perspective of the discipline has not yet been created.
1.2. Objectives
This white paper addresses the five elements of a military theory: define, categorize,
explain, connect, and anticipate. In the areas of “explain” and “anticipate”, the focus is
on identifying and characterizing key “rules of thumb” and principles for cyber
elements.
The scope of the white paper is restricted in two key areas. First, we focus
attention on the national security domain. Changes in cyberspace are having a major
affect on social, cultural, and economic issues, but we address them only tangentially.
Second, we limit attention to the key cyberpower issues that are confronting the
national security policy maker. Thus, there is no attempt to generate a comprehensive
theory of cyberpower that touches on broader issues.
1.3. Approach
The preliminary theory of cyberpower emerged from three initiatives. First, we drew
insights from observations of cyber events, experiments and trends. Second, we
extrapolated from prior national security methods, frameworks, theories, tools, data,
and studies, which were germane to the problem. Finally, we formulated and
hypothesized new methods, frameworks, theories, and tools to deal with unexplained
trends and issues.
Subsequently, the theory has evolved based on two activities. First, over the past
year, several conferences and workshops have been convened that focused on three key
issues: cyber-deterrence, risk management, and international perspectives on cyber
issues. Second, a number of papers have been generated that focused on assessing
cyber policy issues and the US government’s use of social networks. Accordingly, the
preliminary theory has been evolved to reflect the insights that emerged from those
activities.
1.4. Structure
This white paper has adopted the holistic cyber framework depicted in Figure 1. This
framework is patterned after the triangular framework that the military operations
research community has employed to decompose the dimensions of traditional warfare.
In that framework, the base consists of systems models, upon which rests more
complex, higher orders of interactions (e.g., engagements, tactical operations,
campaigns).
Figure 1. Broad Conceptual Framework

By analogy, the bottom of the pyramid consists of the components, systems, and
systems-of-systems that comprise the cyber-infrastructure. The output from this cyber-
infrastructure enhances the traditional levers of power: political/diplomatic,
informational, military and economic (P/DIME). These levers of power, in turn,
provide the basis for empowerment of the entities at the top of the pyramid. These
entities include, inter alia, individuals, terrorists, trans-national criminals, corporations,
nation states, and international organizations. Note that while nation states have access
to all of these levers of power, the other entities generally have access to only a sub-set
of them. In addition, initiatives, such as deterrence and treaties, may provide the basis
for limiting the empowerment of key entities.
The pyramid suggests that each of these levels is affected by institutional factors.
These include factors such as governance, legal considerations, regulation, sharing of
information, and consideration of civil liberties.
It must be emphasized that this framework is merely one of many frameworks that
could be constructed to conceptualize the cyber domain. However, it has proven useful
in decomposing the problem and developing subordinate frameworks to address key
cyber issues.
1.5. Key Definitions
As noted above, there is a continuing discussion about the appropriate definitions for
key cyber terms. In the definition posed by William Gibson, in his 1984 book
“Neuromancer” [3] cyberspace was characterized as: “A consensual hallucination… A
graphic representation of data abstracted from banks of every computer in the human
system.”
For the purposes of this theory, this white paper has adopted the formal definition
of cyberspace that the Deputy Secretary of Defense formulated: “…the interdependent
network of information technology infrastructures, and includes the Internet,
telecommunications networks, computer systems, and embedded processors and
controllers in critical industries” [4]. This definition does not explicitly deal with the
information and cognitive dimensions of the problem. To deal with those aspects
explicitly, we have introduced two complementary terms: cyberpower and
cyberstrategy.
This white paper has adopted the following definition for the term “Cyberpower”.
It is “the ability to use cyberspace to create advantages and influence events in the
other operational environments and across the instruments of power.” In this context,
the instruments of power include the elements of the P/DIME paradigm. For the
purposes of this evolving theory, primary emphasis will be placed on the military and
informational levers of power.
Similarly, the term “Cyberstrategy” is defined as “the development and
employment of capabilities to operate in cyberspace, integrated and coordinated with
the other operational domains, to achieve or support the achievement of objectives
across the elements of national power.” Thus, one of the key issues associated with
cyberstrategy deals with the challenge of devising “tailored deterrence” to affect the
behavior of the key entities empowered by developments in cyberspace.
One of the major issues associated with cyberspace is the question of whether it is
“…an operational domain…”. To explore this issue, note that the term “domain” is not
defined formally in key national security and military products. However, it is cited in
selected policy documents. For example, the 2004 National Military Strategy [5] states
that “The Armed Forces must have the ability to operate across the air, land, sea, space,
and cyberspace domains of the battlespace”. Furthermore, in the 2006 Quadrennial
Defense Review (QDR), it notes that “The DoD will treat cyberspace as a domain of
warfare”.
Joint Publication 3-0 [6] identifies several key features of a domain: it can be
described physically; there are distinctions in means, effects, and outcomes; and
military and combat operations can be conducted in and through the domain.
One can make the argument that cyberspace is a domain through the following
logic. It is widely accepted that (outer) space is a domain. In comparison to “space”,
“cyberspace” has the following bounding attributes that suggest that it is a military
domain. It is subject to ongoing levels of combat; it is characterized by greater ease of
access; and it is more difficult to identify and track military operations within it.
If cyberspace is a domain, it has significant practical implications. It will require

the allocation of resources to support organization, training, and equipping of “cyber-
forces”. It implies the need to develop a culture that is consistent with cyber activities.
Finally, it has implications in the development of a professional cadre and
establishment of a structured career progression. Thus, for the purposes of this evolving
theory, it will be assumed that cyberspace is “an operational domain”.
Consistent with this white paper’s definition, the elements of the holistic
framework can be recast as depicted in Figure 2.
Figure 2. Cyberspace, Cyberpower, Cyberstrategy, and Institutional Factors
1.6. Required Intellectual Capital
To deal with the rich array of cyber policy issues that confront senior decision makers,
it will require a diverse set of intellectual capital. Figure 3 suggests the differing types
of knowledge that will be required to address issues within and across the categories of
interest.
For example, in the realm of cyberspace, there is a need for physicists, electrical
engineers, computer scientists, systems engineers, and system-of-system engineers.
These professionals will play key roles in developing the hardware components (e.g.,
microprocessors, hard drives), software protocols and standards (e.g., implementing
Internet Protocol version 6 (IPv6)), applications and services, and the systems that
exploit this hardware and software (e.g., command, control, and communications
systems).
In the realm of cyberpower, there is a need for subject matter experts (SMEs) that
are qualified to deal with issues of Politics, Diplomacy, Information, Military, and
Economics. This implies extensive reliance on economists (both micro- and macro-)
and social scientists with training in such diverse fields as sociology, cultural
anthropology, psychology, and demographics.Furthermore, in the area of military
knowledge, there is a need for participation by military planners, operators, and
analysts.
In the realm of cyberstrategy, there is a need for interdisciplinary experts who are
able to deal with the full range of political, military, economic, social, informational,
Figure 3. Required Intellectual Capital
and infrastructure (PMESII) issues associated with entities that are empowered by
changes in cyberspace. In particular, analysts are needed who have had experience in
addressing deterrence among these entities.
Finally, in the realm of institutional factors, the key skills required are legal,
governance, civil liberties, and industrial experience.
It is anticipated that one of the main users of this intellectual capital will be cyber
policy decision makers. They will also need operations analysts to help them
orchestrate and harness this heterogeneous intellectual capital, and futurists to help
them conceptualize possibilities that require unfettered imaginations.
2. Theoretical Perspectives
Three of the major objectives of a theory of cyber are to help explain, connect, and
anticipate key aspects of the problem to the decision maker. To do so, it will require
the formulation of conceptual models for the various categories introduced above. In
formulating these conceptual models, it is useful to recall the famous epithet from the
statistician George Box: “all models are wrong; some are useful” [7]. The challenge for
the theorist is to suggest and apply appropriate models that are useful for the decision
maker, and to delineate the range of their utility.
This section systematically introduces a variety of conceptual models that are
germane to the many policy questions associated with cyber issues. Structurally, we
will pursue a “bottom-up” approach and address cyberspace, cyberpower,
cyberstrategy, and institutional factors. For each area, we will introduce a variety of
models and frameworks that will help the decision maker explain key observables and
conceptualize the issues of interest. This will be followed by articulating key “rules of
thumb” and principles that highlight major issues of interest.
2.1. Theoretical Aspects of Cyberspace
This section of the white paper identifies key trends in cyberspace and discusses
cyberspace “rules of thumb” and principles.
2.1.1. Key Trends

This section of the white paper briefly explains key trends in cyberspace. Trends are
introduced in five key areas: growth in users, features of key components (e.g.,
microprocessors, hard drives), architectural features (e.g., Internet Protocols), and
military systems-of-systems.
2.1.1.1. Growth in Users
The most remarkable aspect of the Internet has been the exponential growth in users,
world-wide. Figure 4 illustrates that growth over a thirty-three year period. It can be
seen that the user population increased from approximately 1M users in 1992 to
1,200M users in 2007. It is projected that the Internet will have 2B users by 2010. This
number is projected to grow substantially if the One Laptop Per Child (OLPC) project
is brought to fruition. That project aims to get many millions of low-cost laptops in the
hands of children in under-developed countries.
Figure 4. Number of Internet Users (Millions)
The Navy’s Special Studies Group depicted this growth from another perspective.
They used 50M users as a benchmark for penetration of a mass medium. That level was
achieved by radio in 38 years, television in 13 years, and the Internet in 6 years
(beginning with the introduction of the World Wide Web).
Another key element of cyberspace is cellular telephony. As a point of reference,
the first cell phone call was made in 1973. It is estimated that today, thirty five years
later, approximately 3.3B cell phones are in use, world-wide.
Two other benchmarks serve to calibrate the problem. It is estimated that around
210B e-mails were sent every day in 2008. That is equivalent to 2M e-mails sent every
second. It is estimated that on the order of 70 percent of these e-mails may be spam or
viruses.
Figure 5. Hard Drive Capacity
2.1.1.2. Components
From a theoretical perspective, the physics of the hardware that supports cyberspace
has a significant impact on its performance. This is particularly manifested in the
design of microprocessors and hard drives.
2.1.1.2.1. Microprocessors
Clock cycles of modern microprocessors exceed 4 GHz. Therefore, under ideal
circumstances, electrons can move a maximum of 0.075 meters in a single processor
clock cycle, nearing the size of the chip itself. With clock cycles going even higher 2,
electronic signals cannot propagate across a chip within one clock cycle, meaning
elements of the chip cannot communicate with other elements on the other side of the
same chip. Thus, this limitation maximizes the effective size of a single integrated
microprocessor running at high clock speeds. Addressing this limitation is one of the
reasons that various processor manufacturers have moved chip architectures toward
multi-core processors, where multiple, semi-independent processors are etched on a
single chip. Current chips typically have two or four cores although there are instances
where 1000 to 4000 cores are a single die.
2.1.1.2.2. Hard Drives
Figure 5 depicts computer hard drive storage capability (in gigabits per square
centimeter) over the last twenty five years. It is notable that the improvement in
memory was marginal for the first twenty years until IBM engineers applied the
phenomenon of giant magnetoresistance3. Currently, improvements in memory are
2
As a bounding case, note that in 2008 the fastest US computer, Roadrunner (built by IBM and Los Alamos
National Laboratory), was capable of more than 1 petaflop (i.e., 1 quadrillion floating point calculations per
second) [8].
3
The Nobel Prize in Physics for 2007 was awarded to Albert Fert, France, and Peter Grunberg, Germany,
who independently discovered this phenomenon.
manifesting exponential improvement, making it feasible to create very portable

devices, such as iPods, with extremely high storage capability.
These two examples suggest that a careful technology assessment is needed to
assess if and when bottlenecks in technology will be overcome that limit current
performance.
2.1.1.3. Architectural Features
Figure 6 schematically depicts the architecture of the existing Internet. The key
innovations of this architecture revolve around the key protocols and standards
instantiated in the Transmission Control Protocol/Internet Protocol (TCP/IP) stack and
the use of a router to transmit packets from the sender to the user.
Figure 6. Protocol Layering and Routing Packets Across a Network
Originally, this architecture was devised by a group of colleagues for whom

security was a secondary issue. Thus, the primary emphasis was to implement an
architecture that facilitated the interoperability among heterogeneous networks. In
addition, a decision was made to implement IP addresses that consisted of 32 bits (or
approximately 4 billion addresses).
These two decisions have led to several major limitations in the current
architecture. In light of the security shortfalls in the existing architecture, there is
interest in alternative architectures that are designed around different priorities (e.g.,
highest priority: security; second priority: connectivity among highly mobile users).
Consistent with those revised priorities, new architectural efforts are underway at the
National Science Foundation and DARPA.
Second, the constraint on IP addresses (as well as concern about enhanced security
and mobility) has led to the adoption of IPv6. Since it allocates 128 bits to IP
addresses, it will give rise to an extraordinarily large number of IP addresses 4.
Both of these innovations pose a problem to the cyberspace community: how can
one transition from the current architecture to an alternative architecture, efficiently and
4
IPv6 will provide 2128 addresses. This would provide 5x1028 addresses for each of the 6.5B people alive
today. Alternatively, our sun converts 3.4e38 hydrogen nuclei to helium nuclei each second. Each hydrogen
atom could have its own IPv6 address.
effectively, without creating new security vulnerabilities? This is an on-going challenge

that the computer science community must confront over the next decade.
2.1.1.4. Military Systems-of-Systems
The military community has embraced the underlying computer science principles
associated with the Internet, although they have enhanced security for classified
systems by developing “air gapped” networks (e.g., Secure Internet Protocol Router
Net (SIPRnet), Joint Worldwide Intelligence Communications System (JWICS)).
Figure 7 provides a cartoon of that implementation for the notional Global Information
Grid (GIG).
Figure 7. A Framework to Characterize the GIG
There are several distinctive aspects of the evolving GIG. First, for the transport
layer, the plan is to employ a heterogeneous mix of satellite (e.g., Advanced Extremely
High Frequency), airborne (e.g., selected Joint Tactical Radio Systems (JTRS)), and
surface (e.g., fiber optic) telecommunications media. As a side note, the military is
finding it difficult to develop many of these elements within acceptable levels of
performance, schedule, and cost.
Second, there is interest in employing a Service Oriented Architecture (SOA) to
provide loose coupling among key systems. Third, they have developed Communities
of Interest to address the challenges associated with the data that will flow through the
systems (e.g., specify metadata; deal with issues of pedigree). It has been articulated
that they wish to transition from the principle of “need to know” to “need to share”.
Finally, they hope to assimilate the Services’ visions of future systems into the GIG
(e.g., USA LandWarNet; USN ForceNet; USAF C2 Constellation).
In order to achieve this vision it will require the concerted efforts of the military’s
system-of-systems engineers [9].
2.1.2. Cyberspace “Rules of Thumb” and Principles

To help explain the various trends in cyberspace, one can provide several “rules of
thumb” and strawman principles. Several “rules of thumb” are employed in the
community that are incorrectly characterized as “laws”. For example Moore’s “Law”
indicates that the number of transistors on a chip approximately doubles every 18
months5. This has contributed to the production of devices that have decreased cost,
enhanced computational power, and decreased size. Although this trend is generally
representative of past behavior, there is concern that it may be extremely difficult to
sustain that trend in the indefinite future without a fundamental, expensive change in
the underlying technology (e.g., transition to nanotechnology). Second, as noted above
in Figure 5, recent break-throughs in physics have put the growth in hard drive capacity
on an exponential curve, vice a conservative linear curve.Ultimately, this curve will
reach a level of saturation (an “S-curve”) that is representative of a mature technology.
Lastly, the current limitation in IP addresses will be dramatically overcome once the
transition to IPv6 is implemented.
Based on prior cyber research activities, several strawman cyberspace principles
can be articulated. First, the offensive has the advantage. This is due, in part, to the
“target rich” environment that an adversary faces. This makes it difficult for the
defense to prioritize and defend selected targets. In addition, the existing architecture
makes it very challenging to attribute an attack if an adversary seeks to be anonymous.
If cyberspace is to be more resistant to attack, it will require a new architecture that has
“designed in” security. However, it will be a challenge to transition, effectively and
efficiently, from the current legacy system to a more secure objective system.
2.2. Theoretical Aspects of Cyberpower
This section of the white paper briefly explains key trends in the military and
information dimensions of cyberpower. It focuses on changes in environmental theories
of power and risk, net-centric operations (NCO), and the mission-oriented approach to
influence operations.
2.2.1. Environmental Theories of Warfare

In the discussions that led to this study, it was observed that the naval theories of
Alfred Mahan played a major role in shaping the US perspectives and strategies on
naval power. It was suggested that cyber power needed a comparable perspective to
shape its strategy in cyberspace.
Consistent with that interest, this study re-evaluated the various environmental
theories of power. These included analyses of land power (Mackinder [11]), naval
power (Mahan [12]), airpower (Douhet [13]), and space power (Gray and Sloan [14]).
Based on these analyses, four common features of environmental power theories were
identified: technological advances; speed and scope of operations; control of key
features; and national mobilization.
5
To put this change in context, note that in 1971, processor speeds were on the order of 4x105 Hertz (or
400 KHz) and the cost of 1 MB of Dynamic Random Access Memory (DRAM) was approximately $400 (in
2006 dollars). By 2006, commercial processor speeds were on the order of 4x109 Hz (or 4 GHz) and the cost
of 1 MB of DRAM was $0.0009 [10].
Consistent with each of these features, the following implications were drawn for a
theory of cyberpower. With respect to technological advances, it was observed that
dependency on cyberspace has given rise to new strategic vulnerabilities. This
vulnerability has been dramatized by the specter of a “cyber Pearl Harbor” and the
realization that the existing cyberspace is vulnerable to a variety of adversary attacks
(e.g., denial of service attacks, exfiltration of sensitive but unclassified information;
potential corruption of sensitive data). In addition, due to the diffusion of low cost
cyberspace technology, the power of non-states (e.g., individuals, terrorists,
transnational criminals, corporations) has been greatly enhanced (see below).
Improvements in cyberspace have also served to enhance the speed and scope of
operations. This is manifested in the speed at which global operations can be conducted
(e.g., the ability to successfully engage time sensitive targets, any where in the world).
In addition, it has led to improvements in the ability to automate command and control,
dramatically decreasing the classic Observe-Orient-Decide-Act (OODA) loop process.
In the environmental theories of power, emphasis was placed on controlling key
features. For example, in naval theories this entailed the control of key “choke points”
(e.g., the Straits of Malacca), while in space power, there was interest in controlling
key geosynchronous orbit locations. In the case of cyberspace, the key features of
interest are man-made. Thus, for example, there is interest in defending “cyber hotels”
where key information and communications technology (ICT) systems are
concentrated. In addition, while the choke points in the physical world tend to be
immutable, they may change relatively rapidly in cyberspace (e.g., location of
extensive server farms).
Finally, national mobilization is a key measure of cyberpower. To ensure that it is
available when needed, it is vital to ensure that the US has access to a cadre of
cyberspace professionals. This argues for re-examining career progression for
cyberspace professionals in the military Services. In addition, it is important to
establish links to the private sector where the bulk of cyberspace professionals reside.
This suggests that a reservoir of reservists should be established to provide access to
this intellectual capital in the event of national need.
It is argued in this white paper that the US Government (USG) has tended to focus
on the opportunities offered by changes in cyberspace, rather than the risks that we are
assuming. To summarize that dichotomy, Table 1 identifies the opportunities and risks
associated with military activities at the strategic, operational, and tactical levels.
As can be seen in Table 1, the risks at the strategic level include loss of technical
advantage (due to the diffusion of cyberspace technology), potential rapid change in the
operating environment (e.g., possibility that nations such as China could “leap-frog”
the US by transitioning rapidly to IPv6), and the vulnerabilities associated with military
dependence on key systems (e.g., the GIG). At the operational level, the diffusion of
cyberspace technology could result in the US loss of advantage in operational pace.
Finally, at the tactical level, advances in cyberspace could generate a new front for
adversaries to build resources. These observations suggest that the USG might be
assuming significant, unknown risks by failing to take a balanced perspective of key
cyberspace trends. It also implies the need to undertake more extensive risk
assessments to understand the potential “down-side” of key dependencies.
Table 1. Military Opportunities & Risks in Cyberspace
To begin to deal with these risks, steps should be taken at the strategic, operational,
and programmatic levels. At the strategic level, steps should be taken to ensure the
resilience of supporting critical infrastructures (e.g., electric power generation and
transmission). At the operational level, it is vital to plan to conduct operations against
an adversary that is highly cyberwar-capable. This should include the creation of a
highly-capable Opposing Force (OPFOR) that would be employed extensively in
experiments and exercises. Finally, at the programmatic level, emphasis should be
placed on addressing cyberspace implications in the development process. This should
include placing higher priority on the challenges of Information Assurance. Overall, an
improved analytic capability is required to address each of these issues.
2.2.2. Net-Centric Operations (NCO)

As one aspect of the analytic capability, work is needed to enhance and apply the
existing conceptual framework for NCO. As illustrated in Figure 8, the NCO process
involves consideration of the interactions among the physical, information, cognitive,
and social domains6. There is a need to develop better analytic tools for all aspects of
this process, particularly in the cognitive and social domains. One potential source of
intellectual capital is the ongoing initiative by the Director, Defense Research and
Engineering (DDR&E), OSD, to improve human, social, cultural behavior (HSCB)
models and simulations.
2.2.3. Mission Oriented Approach to Influence Operations

In the area of influence operations, a strawman framework has been developed to help
the community plan for and implement influence operations (Figure 9). This
framework represents an extension of the Mission Oriented Approach to Command and
Control (C2) that was developed and applied to a variety of C2 issues in the 1980s [16].
This approach begins with the articulation of the nature of the problem of interest.
It then poses a sequence of questions. First, what is the operational objective of the
6
Note that the figure does not explicitly depict the social domain.
Figure 8. Conceptual Framework for NCO
operation? A reasonable objective may be to establish a trust relationship with the

indigenous population (vice “winning their hearts and minds”) [17]. Second, how
should this operational objective be accomplished? Again, a decision was made to
work with surrogate audiences in order to reach the undecided population. These
surrogate audiences included the local media, religious leaders, educational leaders,
political leaders, and tribal leaders. Consistent with those surrogate audiences,
organizations and processes were established to reach out to them effectively. At this
point, one can characterize the existing Doctrine, Organization, Training, Materiel,
Leadership and Education, Personnel, and Facilities (DOTMLPF) activities and
compare them to the operational needs. This will give rise to DOTMLPF shortfalls and
the articulation of options to mitigate them. It may also prompt the operator to re-
evaluate the operational goals and the operational activities to support them.
This process should be refined and applied to a broader variety of strategic,
operation, and tactical influence operations. In particular, it can be used to explore the
utility of employing new options in cyberspace to improve future influence operations
(e.g., the “New Media”, such as the Internet and social networks).
2.2.4. Cyberpower “Rules of Thumb” and Principles

One of the so-called “laws of cyberpower” was formulated by Bob Metcalfe [18]. He
postulated that the value of a telecommunications network is proportional to the square
of the number of users of the system (n2). However, there is no empirical data to
support this “law”. In a recent article [19], it is observed that the value 7 is closer to
nlog(n).
7
To illustrate the differences in these results, assume that one has a network of 100 users. According to
“Metcalfe’s Law”, the “value” of the network is on the order of 10,000. However, the revised “Law”
suggests that the value is on the order of 100x2=200.
Figure 9. Strawman Framework for Analyzing Influence Operations
From an analytical perspective, the former Office of Force Transformation has

supported a number of studies to relate the impact of net-centricity on enhancements in
cyberpower (primarily in the military domain). These studies have demonstrated that
net-centricity can have a substantial affect on mission effectiveness for selected
mission areas. For example, the use of Link 16 by airborne interceptors in M-on-N
combat can enhance air-to-air loss exchange ratios by approximately 2.5 [20].
However, the complexity of modern conflict is such that it is difficult to assess the
affect of net-centricity on complex missions (e.g., air-land operations; stability and
reconstruction operations). This suggests that additional experiments will be needed to
assess the quantitative value of net-centricity for complex missions, in which better
control is exercised over potentially confounding variables.
2.3. Theoretical Aspects of Cyberstrategy
This white paper has identified an extensive list of entities that are being empowered
by changes in cyberspace. This list includes individuals, hacktivists 8, non-governmental
organizations (e.g., Red Cross), terrorists, trans-national criminals, corporations,
nation-states, and international governmental organizations (e.g., the United Nations).
For the purposes of this white paper, attention has been focused on a sub-set of
these entities. These include terrorists, trans-national criminals, and a subset of nation
8
Wikipedia definition: Hacktivism (a portmanteau of hack and activism) is often understood as the writing
of code, or otherwise manipulating bits, to promote political ideology…
states (e.g., Estonia, China, Russia). From a USG national security perspective, two key
issues stand out. First, is it feasible to achieve “tailored cyber deterrence”? Second,
what steps should be taken to deal with cyber espionage?
2.3.1. Terrorist Use of Cyberspace

Terrorists are being empowered substantially by changes in cyberspace. With the loss
of physical sanctuary in key areas (e.g., Afghanistan), they have been turning to the
sanctuary of cyberspace to perform a variety of key, inter-related functions. These
functions include, inter alia, recruiting of malleable candidates, raising resources to
support their operations, planning their operations (employing such open-source tools
as Google Earth), commanding and controlling their operations, conducting influence
operations (e.g., disseminating their perspectives of operations in Iraq to sympathetic
and uncommitted audiences), and educating and training supporters on a variety of
subjects (e.g., interpretations of the Koran; building and deploying Improvised
Explosive Devices (IEDs)).
Terrorists have found cyberspace to be an attractive milieu for several reasons.
First, the cost of entry is low. One can acquire the latest cyber technology for hundreds-
to-thousands of dollars and exploit key open-source software. In addition, terrorists can
take full advantage of the extraordinary sums that have been invested by the
commercial sector in cyber infrastructure (including communications and navigation
systems). Second, cyberpace provides rapid, world-wide reach. Thus, they are able to
transcend the limited geographic reach of their prior physical sanctuary and perform the
key functions cited above. Third, it has been posited that the next-generation terrorists
are being radicalized by on-line interactions[21]. Finally, there is concern that terrorists
are developing linkages with trans-national criminals to support their objectives. The
trans-national criminals are able to provide terrorists with cyber knowledge while
profiting from the relationship.
Recently, a number of reports have been issued that suggest strategies for the USG
to pursue to counter the terrorists’ use of cyberspace. As an illustration, the Special
Report on Internet-Facilitated Radicalization [22] formulated five recommendations to
address the cyber threat posed by terrorists9. The many actions associated with those
recommendations are summarized in Table 2. From the perspective of this white paper
on cyberspace theory, some of the more interesting actions involve developing a
strategic communication plan based on a compelling narrative, implementing an
innovative program on behavior science research, and addressing USG shortfalls in
knowledge of culture and language.
9
This report recommended that five steps be taken:
• Craft a compelling counter-narrative for worldwide delivery, in multimedia, at and by the grassroots level.
• Foster intra- and cross-cultural dialogue and understanding to strengthen the ties that bind together
communities at the local, national, and international levels.
• Recognize and address the need for additional behavioral science research into the process of radicalization
both online and offline.
• Deny or disrupt extremist access to, and extremist efforts through, the Internet via legal and technical
means, and covert action, where appropriate.
• Remedy and resource capability gaps in government.
Table 2. Options to Counter Terrorist Use of Cyberspace
2.3.2. Criminal Use of Cyberspace

At a recent workshop on cyber issues at CTNSP, several of the participants focused on
the challenges posed by cyber crime.Several of the speakers and panelists emphasized
that the threat is real (and expanding). The speakers stated that “We are losing the
global cyber war at an accelerated rate.” In addition, they stated that “Cybercrime is
effective because you can try to commit crimes an infinite number of times, but you
need to succeed only a few times.” Overall, it was stated that there are three elements
of the threat: crime; industrial espionage; and traditional espionage. It was further noted
that criminal attack vectors are comparable to those of state attacks.
The speakers also made the following observations. It was recommended that we
focus on the “top 25 Common Weaknesses Enumeration (CWEs)”. Furthermore, many
of the panelists observed that current laws to deal with these issues provide limited
value. In addition, it was noted that Web developers and code writers have no idea how
to write secure code.
This raised the following question: When will the tide turn? It was suggested that
we will make useful headway when we implement the following steps. First, it is
critical to create safer software. One recommendation was to make business
partnerships contractual (e.g., require the company to fix future flaws and security
problems in the software). Second, it was observed that we need to stop existing attacks
(e.g., implement more effective actions by the US Department of Justice and computer
security specialists). However, in order to do so, we need to find the needed talent. As
an example, it was observed that China’s People Liberation Army periodically runs
national talent searches for the best hackers.
2.3.3. Nation State Use of Cyberspace

From a nation-state perspective, different combinations of levers of power are
employed to generate desired effects. From a theoretical perspective, these nations
formulate their strategy though a mix of P/DIME activities. The effects of these
activities are manifested in the areas of PMESII. Tools are being created to explore
how alternative P/DIME activities can give rise to differing PMESII effects (see
discussion below).
2.3.3.1. United States Use of Cyberspace
Using the P/DIME-PMESII paradigm, one can begin to characterize how cyber
changes have empowered the US. In the political dimension, changes in cyberspace
have encouraged democratic participation by the population. With respect to the
Internet, it has provided a forum for the individual to articulate his views (e.g.,
proliferation of blogs, contributions to wikis). In addition, political candidates are
finding the Internet to be a useful vehicle for raising resources from grass root
supporters. Furthermore, Internet sites such as YouTube have enhanced the
accountability of candidates.
In the military dimension, the concept of NCO has enhanced effectiveness in
selected operational domains (e.g., air-to-air combat). Efforts are still required to
quantify the military benefits that are achievable for more complex military operations
(e.g., air-land maneuver).
Economically, the commercial sector has seen dramatic improvements in industrial
productivity (e.g., Boeing’s use of computer aided design tools to support the
development of the 777 aircraft and the more recent development of the 787). These
cyber-based advancements are giving rise to considerable improvements in
responsiveness (e.g., time to market) and cost reductions (e.g., outsourcing “back-room
operations” to other nations).
Socially, the development of cyberspace has increased social interactions in
several ways. Tens of millions of users participate in social networking sites (e.g.,
MySpace, FaceBook). In addition, millions of users, world-wide, participate in virtual
reality environments (e.g., Second Life). In fact, it has been rumored that terrorist
organizations are using virtual reality environments to explore proto-typical operations.
In the information dimension, the Internet has increased dissemination of
information, world-wide. Given the US’ strong position in entertainment (movies,
games) and advertising, it is argued that it provides a strong forum for promoting “soft
(or smart) power” [23].
Finally in the infrastructure dimension, many critical infrastructures have been
using the Internet to facilitate more efficient and effective operations. However, this
constitutes a “double edged sword” because of the potential vulnerability of
Supervisory Control and Data Acquisition (SCADA) systems [24].
Overall, it must be stressed that empowerment is more than the sum of the
individual PMESII factors.
2.3.3.2. Near-Peer Use of Cyberspace
Various studies of nation-state empowerment provide insights on the projected uses
and cyber-strategies of China and Russia. The relevant white papers discuss the recent
writings from key conceptual thinkers in those nations and compares and contrasts
these strategies. Those nations use a different vocabulary in discussing cyberspace and
cyberpower. For example, Chinese writings on the subject focus on stratagems,

objective and subjective reality, and the dialectic 10.
Two key aspects of the Chinese view of the Revolution in Military Affairs are
particularly germane: “War with the objective of expanding territory has basically
withdrawn from the stage of history, and even war with the objective of fighting for
natural resources is now giving way to war with the objective of controlling the flow of
financial capital.”
Furthermore: “If we go our own path to develop military theory, weapons, and
equipment, we will develop something never seen before in places that no one has ever
thought of before; others will be unable to anticipate or resist our ‘self-accommodating
systems’.”
As an illustration of “self-accommodating systems” against the superior foe, three
ways are cited for making a cat eat a hot pepper: “stuff it down his throat, put it in
cheese and make him swallow it, or grind it up and spread it on his back. The latter
method makes the cat lick itself and receive the satisfaction of cleaning up. The cat is
oblivious to the end goal. This is strategy.”
2.3.4. Cyber Deterrence

There are three key challenges that must be addressed to deal with cyber deterrence
[25]. These include the challenges of attribution, the lack of a cyber-deterrence track
record, and the occurrence of unexpected higher-order effects.
The primary challenge is the perceived difficulty of attributing such attacks to a
specific attacker (e.g., state, non-state actor). Note, for example, if competitors believe
we cannot determine who is attacking us in cyberspace, they may convince themselves
that such attacks involve little risk and considerable gain. However, there is a key
trade-off that must be weighed: if we demonstrate our ability to detect and attribute
cyberspace attacks we may provide intelligence about our capabilities. Thus, we may
be posing a greater cyberspace threat to the nation in the future.
The second key challenge is the lack of a known historical track record of US
detection, attribution, and response. This poses a series of key issues. They include the
credibility of deterrent actions, emboldening potential attackers, and defining publicly
what the US considers a cyberspace “attack” and the potential kinds of responses to
such attacks.
The third challenge is the potential for producing higher order effects that might
result in unintended consequences and possibly undesired consequences. There are
three key issues associated with that challenge. First, it is a function of the nature of the
attacker’s goals and objectives. Second, if the competitor is concerned about
unintended consequences, it could enhance the effects of our deterrence activities if it
wishes to control escalation or fears “blowback” from its cyberspace operations.
Finally, if the competitor’s goal is to create chaos, deterrence could be undermined by
the potential for unintended consequences.
In addition, there is interest in “tailoring deterrence” [26] to address the variety of
adversaries that exist in cyberspace (e.g., non-state actors; state actors). However, there
is a debate within the analytic community as to whether tailored deterrence is a viable
concept for the full spectrum of adversaries of the US [27]. That issue represents an
important element of the research agenda for the community. However, it is
10
“Reasoning that juxtaposes opposed or contradictory ideas and seeks to resolve conflict”.
hypothesized that the full set of P/DIME options should be considered in developing a
course of action to respond to a cyber attack. For example, the US might respond to a
cyber attack through a variety of levers of power including diplomacy (e.g., a
demarche) or economic actions (e.g., restricting the flow of technology).
2.3.5. Cyberstrategy “Rules of Thumb” and Principles

In weighing the cyberstrategy insights, three key insights emerged. First, the “low end”
users (e.g., individuals, hacktivists, terrorists, trans-national criminals) have enhanced
their power considerably through recent cyberspace trends. A tailored deterrence
strategy will be needed to keep these entities in check.
Second, potential near-peer adversaries are aggressively exploring options to
exploit attributes of cyberspace. In the near term, this is being manifested through acts
of espionage that have resulted in the exfiltration of massive amounts of sensitive
governmental and industrial data [28]. In the longer term, the US must be prepared to
deal with unique “cyber strategems” that reflect the unique cultural and military history
of key nations (e.g., China, Russia).
To deal with the emerging cyber threat, the US must conduct experiments and
exercises that feature a creative and aggressive cyber opposing force. It would be naïve
and dangerous to assume that future adversaries will not seek to negate the benefits that
the US hopes to achieve through net centric warfare.
2.4. Theoretical Aspects of Institutional Factors
This section of the white paper focuses on two critical institutional factors: governance
of cyberspace and the legal dimensions of the problem. The section concludes by
identifying key institutional issues and principles.
2.4.1. Governance
Table 3 characterizes key governance functions in cyberspace and the organizations
that participate in these functions. It can be seen that the mechanisms for governance of
the Internet are exceedingly complex. Organizational activities often overlap or fit end-
to-end, requiring the expenditure of considerable resources in multiple forums to
achieve objectives. Consequently, there is a core set of participants (generally in the
private sector) that are involved in several of these key organizations.
In an effort to evaluate the performance of Internet governance, we have
introduced the following criteria: open, democratic, transparent, dynamic, adaptable,
accountable, efficient, and effective. When assessed against these criteria, one can
conclude that recent Internet governance has performed remarkably well.
However, as we look to future, the USG will be challenged to alter its position on
Internet governance. Preliminary views on this subject are being articulated at the
ongoing Internet Governance Forums (IGF). In fact, a recent white paper on the subject
[29] made the following observations:
“Internet Governance is an isolating and abstract term that suggests a nexus with
an official government entity. The term also implies a role for the US Congress in
Internet decision-making. It is a misnomer because there is no true governance of the
Internet; only a series of agreements between a distributed and loosely connected group
Table 3. Governance of Cyberspace
of organizations and influencers. A more fitting term may be ‘Internet Influence,’ or for
long-term strategy purposes, ‘Internet Evolution’.”
2.4.2. Cyber Law

One of the most challenging legal issues confronting the cyber community is as
follows: “Is a cyberattack an act of war?” Legalistically, the answer is often presented
as one of three possible outcomes: it is not a use of force under UN Article 2(4); it is
arguably a use of force or not; it is a use of force under UN Article 2(4).
There are several frameworks that are being considered by the legal community to
address this issue. Michael Schmitt has formulated a framework that defines and
addresses seven key factors: severity, immediacy, directness, invasiveness,
measurability, presumptive legitimacy, and responsibility [30]. Once one has assessed
each of those factors, one should employ multi-attribute utility theory to weight each of
these factors and come to a determination. An associated challenge is to formulate
responses to that attack that are consistent with the legal tenet of proportional response.
Overall, the area of cyber law is in its infancy. Although there have been
preliminary rulings on sharing of music (e.g., Napster), there are major issues on the
questions of sovereignty, intellectual capital, and civil liberties. These issues will be
major areas for research for the foreseeable future.
2.4.3. Institutional Principles

Based on the insights developed during the course of this study, four major strawman
principles have emerged in the arena of Institutional Factors.
First, given the complexity of the governance mechanisms, one should seek
influence over cyberspace vice governance.
Second, the legal community has barely addressed the key issues that must be
resolved in the cyber arena. For example, considerable research is needed to assess the
following key questions:
• What is an act of (cyber)war?
• What is the appropriate response to an act of (cyber)war?
• What is the appropriate way to treat intellectual property in the digital age?
• How can nations resolve differences in sovereign laws associated with cyber
factors?
Third, there is a need for a framework and enhanced dialogue between champions
of civil liberties and proponents of enhanced cyber security to establish an adequate
balance. Finally, guidance and procedures are required to address the issue of sharing
of cyber information between the USG and industry. This approach should be based on
the concept of risk management.
3. Connections
At the beginning of this white paper, it was noted that one of the reasons for a theory
was the need to connect diverse elements of a body of knowledge. In general, the
community is focusing on the issue of connecting the knowledge within a stratum of
the pyramid. Even though this is challenging, it generally involves communicating
among individuals with a common background and lexicon.
It is far more difficult to have individuals connect across the different strata of the
pyramid. This requires individuals from different disciplines to work effectively
together. In order to do so, it requires a holistic perspective on the Measures of Merit
(MoMs) for cyber issues.
Figure 10 suggests a potential decomposition of the MoMs associated with the
cyber problem. It identifies four linked sets of measures: Measures of Performance
(MoPs), Measures of Functional Performance (MoFPs), Measures of Effectiveness
(MoEs), and Measures of Entity Empowerment (MoEEs). Since this field of endeavor
is still in its infancy, the material is meant to be illustrative and not exhaustive.
MoPs are needed to characterize the key computer science and electrical
engineering dimensions of the problem. A key measure is the amount of bandwidth that
is available to representative users of cyberspace. As the bandwidth increases to the
megahertz/sec range, the user is able to access advanced features such as imagery and
video products. A second key measure is connectivity. For circumstances in which the
cyber-infrastructure is fixed, a useful measure is the percent of people in a country that
have access to the Internet. However, in many military operations, the cyber-
infrastructure and the users are mobile. Under those circumstances, a more useful
measure is the performance of Mobile, Ad hoc NETwork (MANET) users (e.g., their
ability to stay connected). Third, one can introduce measures of the “noise” that
characterizes the cyber-infrastructure. For example, the extent to which the quality of
the Internet is degraded can be characterized by the unwanted e-mail that it carries
(“spam”), which can subsume a substantial subset of the network’s capacity. As an
Figure 10. Measures of Merit
example, it has been estimated that in recent months up to 90% of the traffic on the
Internet is spam [31]. In addition, the integrity of the information is further
compromised by “phishing” exploits in which criminal elements seek to employ the
Internet to perpetrate economic scams. Finally, MoPs can be introduced to characterize
resistance to adversary actions, including denial of service attacks, propagation of
viruses or worms, and illicitly intruding into a system.
It is useful to introduce MoFPs that characterize how successfully selected entities
are able to perform key functions, taking advantage of cyberspace. In the case of the
US military, the concept of net-centricity is to employ advances in cyberspace to
perform essential functions. These include the ability to enhance the performance of
increasing levels of information fusion. Similarly, a basic tenet of net-centricity is to
propagate commander’s intent so that the participants in the operation can synchronize
and self-synchronize their actions.
MoEs are needed to characterize how effective entities can be in their key
missions, taking advantage of cyberspace. In the context of Major Combat Operations,
MoEs are needed to characterize the ability to exploit cyberspace in multiple
dimensions. At one extreme, enhancements in cyberspace have the potential to reduce
the time to conduct a campaign and the casualties associated with the campaign. At the
other extreme, enhancements in cyberspace may substantially enhance Blue loss
exchange ratios and the amount of ground gained and controlled.
From the perspective of cyberstrategy, there is interest is characterizing the extent
to which enhancements in cyberspace can empower key entities. In the case of nation
states, potential MoEEs might include selected PMESII variables. As an example, it
might address the ability to leverage cyberspace to influence a population, shape a
nation at strategic crossroads, and deter, persuade, and coerce an adversary.
Table 4 depicts candidate MoMs that may be employed in future cyber analyses.
Table 4. Selected Measures of Merit
4. Anticipation
From the perspective of the decision maker, the key challenge is to anticipate what will
occur next in the cyber domain and to formulate coherent policy to cope with those
issues. To begin to address that challenge, this section deals with four aspects of
anticipation. First, it identifies key trends that are expected to characterize cyberspace.
Second, it identifies the research activities that should be conducted to address those
trends. Third, it briefly identifies the major policy issues that decision makers will need
to address. Finally, it discusses the assessment needs that must be addressed to support
the formulation and analysis of policy options.
4.1. Cyber Trends
It is extremely difficult to provide quantitative estimates as to how rapidly key trends in

cyberspace will be manifested. Thus, the following should be regarded as a partial,
qualitative list of some of the most significant potential changes.
First, there is an increased move to adoption of IP-based systems. As a
consequence, one can anticipate a convergence of telephone, radio, television, and the
Internet. As one example, there is a dramatic use of Voice over IP (VoIP) (with
attendant security issues) in the area of telephony. Second, we are seeing the
emergence of sensor networks that feature an extremely large number of heterogeneous
sensors. As one manifestation, we are seeing the netting of extremely large number of
video cameras in urban areas, raising issues in the civil liberties community. Third, we
are seeing an inexorable trend towards proliferation of broadband and wireless. An
example of this trend was the plan to have city-wide deployment of Worldwide
Interoperability for Microwave Access (WiMax). However, this trend suggests the
difficulty in predicting when a trend becomes a reality. NEXTEL had made this
objective the key to their strategy; however, they have recently observed that the
technology has not matured sufficiently to implement it in the near-term [32]. Fourth,
we are observing enhanced search capabilities, both for local systems and the entire
Internet. One of the keys to this trend has been industrial competition to develop
improved search engines (in part, to enhance advertising revenue). Fifth, we are seeing
extraordinary efforts to enhance human/machine connectivity. As one example, we are
seeing direct nerve and brain connections to computers or prostheses, arising from
efforts to treat soldiers injured by IEDs in Iraq [33]. Sixth, we are seeing dramatic
increases in user participation in information content. This trend is manifested through
the proliferation of video blogs, contributions to wikis, participation in social networks
(e.g., MySpace, FaceBook), and involvement in virtual reality environments (e.g.,
Second Life). Finally, some experts have postulated that we are entering the third phase
of the Internet (i.e., phase 1: communicating; phase 2: content; phase 3: collaboration).
This third phase is characterized by “cloud computing” where “information is stored
and processed on computers somewhere else” [34]. One of the major issues associated
with this paradigm is our ability to provide adequate security for the “cloud”.
4.2. Opportunities for Cyber Research
As an application of the emerging theory of cyber, Table 5 identifies the major areas
where cyber research should be pursued.
Table 5. Areas Where Additional Theoretical Research Is Required
4.2.1. Cyberspace Research

In the area of cyberspace, improved technology projections are needed to identify key
breakthroughs that may substantially affect MoPs for cyberspace. Second, it is
inevitable that malevolent actors (e.g., insiders, adaptive adversaries) will gain access
to the USG and defense industrial base cyberspace. This suggests that research is
needed to protect the essential data in cyberspace from exfiltration or corruption.
Finally, additional research is needed to formulate an objective architecture for

cyberspace that is inherently more secure than the existing architecture. Consistent with
that effort, there is a need to address the challenging issue of transitioning from the
existing to the objective architecture.
4.2.2. Cyberpower Research

Due to resource constraints, this evolving assessment of cyber theory has not
adequately addressed all the levers of power (e.g., political, diplomatic, economic). As
an initial step, assessments should be completed for these other levers of power.
Second, existing assessments of the military lever of power have focused almost
exclusively on the potential benefits that can accrue by creatively employing
cyberspace. It is equally important to perform risk assessments to understand the
potential downside of relying extensively on cyberspace. This includes conducting
experiments and developing the methodology, tools, data, and intellectual capital
required to perform military risk assessments. Similarly, it is important to conduct
research into the potential benefits and risks associated with leveraging cyberspace
developments for non-US military capability (e.g., NATO allies that are pursuing
Network Enabled Capabilities (NEC)). Finally, in the area of information, additional
research is needed to quantify the information duels that are likely to occur with
potential adversaries.
4.2.3. Cyberstrategy Research

To deal with the challenges posed by the full array of entities empowered by
enhancements in cyberspace, it is vital that the information-enabled societies conduct
research on “tailored deterrence”. This concept suggests that key alliances, such as
NATO, must develop a holistic philosophy that understands each of the potential
adversaries (e.g., its goals, culture, risk calculus), develops and plans for capabilities to
deter these adversaries, and develops a strategy to communicate these concepts to the
4.2.4. Institutional Factors Research

Theoretical research is needed to address key gaps in institutional knowledge in the
areas of governance, legal issues, sharing of information, Internet regulation, and civil
liberties.
First, in the area of governance, the USG must reassess the role of the Internet
Corporation for Assigned Names and Numbers (ICANN) in the governance of the
Internet. It is clear that, in the future, the USG must be more adroit in the area of “cyber
influence” vice governance. This will require a thorough re-examination of all the
institutional bodies that affect cyber governance and the development of a USG
strategy to interact with them.
Second, “cyber legal” issues are in their infancy. The current situation is non-
homogeneous with inconsistent laws in various sovereign nations (e.g., German hate-
crime laws; limited signatories to the Council of Europe Convention on Cybercrime
[35]. In particular, there is a need to clarify the issue of espionage in cyberspace (e.g.,
What is it? What rights of response are left to the victims?). In addition, there is a need
to adopt a consistent model that can be applied to determine whether a cyber attack is
an act of war.
Third, there is continued controversy about the sharing of information between the
USG and the private sector. Research is needed to determine what information should
be shared, under what circumstances.
Fourth, it has been observed that regulatory agencies, such as the Federal
Communications Commission, have the authority to regulate Internet Service Providers
(ISPs) to redress selected cyber security issues. However, to date, regulatory agencies
have been reluctant to address these issues.
Fifth, the recent debate about the Foreign Intelligence Surveillance Act (FISA)
court has mobilized the civil liberties community to raise the specter of “Big Brother”.
As a consequence of the actions of civil liberties organizations, key USG programs
have been terminated or modified (e.g., DARPA’s Total Information Awareness (TIA),
DHS’s Multi-state Anti-Terrorism Information Exchange (MATRIX)). Research is
needed to clarify the appropriate balance among actions to deal with adversaries while
still protecting civil liberties.
4.2.5. Cyber Assessment Research

As discussed below, our ability to perform cyber assessments is extremely uneven. As
a consequence, research efforts are required to develop analytical methods, tools, data,
services, and intellectual capital to address key cyber issues in the areas of cyberpower,
cyberstrategy, and infrastructure issues.
Table 6. Selected Policy Recommendations
4.3. Cyber Policy Issues
Several major policy issues have been singled out that require further attention. For the
purposes of this preliminary cyber theory, these issues have served to focus the
boundaries of this study, although we have also addressed a number of lower priority
policy issues. Consequently, emphasis has been placed on assembling the intellectual
capital required to illuminate those issues.
Figure 11. Subjective Assessment of MS&A for Cyber Policy Analyses
In Table 6, these issues have been aggregated into the categories of cyberspace,
cyberpower, cyberstrategy, and institutional factors. Most of these issues are extremely
broad and contentious; consequently, additional analyses will be required to address
them adequately.
4.4. Cyber Assessment
One of the major challenges confronting the analysis community is to develop the
methods, tools, and data needed to support cyber policy decision makers. Figure 11
suggests the relative maturity of key tools in the areas of cyberspace, cyberpower,
cyberstrategy, and institutional factors.
In the areas of cyberspace, there are several tools that the community is employing
to address computer science and communications issues. Perhaps the best know is the
OPNET simulation [36] that is widely employed to address network architectural
issues. From an analytic perspective, techniques such as percolation theory [37] enable
one to evaluate the robustness of a network. Looking to the future, the National
Research Laboratory (NRL) has developed a GIG Testbed to explore the myriad issues
associated with linking new systems and networks.
In the area of cyberpower, the community has had some success in employing live,
virtual, and constructive simulations. For example, in assessments of air-to-air combat,
insights have been derived from the live AIMVAL-ACEVAL experiments, virtual
experiments in the former McDonnell Air Combat Simulator (MACS), and
constructive experiments using tools such as TAC BRAWLER. However, the
community still requires better tools to assess the impact of advances in cyberspace on
broader military and informational effectiveness (e.g., land combat in complex terrain).
In the area of cyberstrategy, a number of promising initiatives are underway. In

response to recent tasking by STRATCOM, a new methodology and associated tools
are emerging (i.e., Deterrence Analysis & Planning Support Environment (DAPSE)
[38]. However, these results have not yet been applied to major cyberstrategy issues. In
addition, promising tools are emerging from academia (e.g., Senturion; GMU’s Pythia)
and DARPA (e.g., Conflict Modeling, Planning & Outcomes Experimentation
(COMPOEX)). However, these are still in early stages of development and application.
Finally, as noted above, there are only primitive tools available to address issues of
governance, legal issues, and civil liberties. Some tools are being developed to explore
the cascading effects among critical infrastructures (e.g., National Infrastructure
Simulation and Analysis Center (NISAC) system dynamics models [39]); however,
they have not yet undergone rigorous validation.
5. Summary
Consistent with the macro-framework that has been adopted to characterize the cyber
problem, this section summarizes the key insights in the areas of cyberspace,
cyberpower, cyberstrategy, and institutional factors. The section concludes by
identifying the next steps that should be taken to refine the theory of cyberpower.
5.1. Key Insights
5.1.1. Cyberspace
Cyberspace is an environment that is experiencing exponential growth in key MoPs.
There is an extraordinary diffusion of knowledge among all the stakeholders of
cyberspace, including malevolent users. As a consequence of this diffusion of
knowledge, cyberspace is being degraded by “noise” (e.g., proliferation of spam) and a
broad variety of cyber attacks. The most troubling of these attacks includes Distributed
Denial of Service, exfiltration of data, and the potential for corruption of data. In each
instance, recent experience has demonstrated that these attacks are relatively easy to
implement (e.g., technically, financially) and extremely difficult to attribute.
These vulnerabilities arise from the basic architecture that has evolved from the
original ARPAnet. A new cyberspace architecture may be required to halt the
perceived erosion of security. However, there will be substantial difficulties in
transitioning from the current architecture to one that is more robust against adversary
action.
5.1.2. Cyberpower
As cyberspace evolves, it has the potential to enhance each of the levers of national
power. This white paper has focused on two of these levers: military and information.
In the area of military power, it was observed that studies are underway to
characterize the extent to which enhancements in cyberspace can enhance key MoEs.
These studies tend to be unambiguous in the area of air-to-air combat where
experiments suggest that enhanced digital communications can enhance loss-exchange
ratios by a factor of approximately 2.5. Although studies of other military operations
have also been undertaken, the results are generally confounded by other factors (e.g.,
mobility, protection).
To complement these experiments, an assessment of theories of environmental
warfare was undertaken that critically reassessed the theories of land, sea, air, and
space theory. Based on that assessment, it was concluded that a theory of cyberpower
should focus on four key factors: technological advances, speed and scope of
operations, control of key features, and national mobilization.
From the perspective of “information”, this white paper has addressed influence
operations from a strategic and tactical perspective. Based on prior experiences and an
adaptation of earlier analytical frameworks, an approach was developed for linking
operational objectives and processes to DOTMLPF requirements. These assessments
suggest that developments in cyberspace can substantially affect future efforts to
enhance influence operations (e.g., implement precision guided messages).
5.1.3. Cyberstrategy
The evolving theory of cyber has identified a range of entities that will be empowered
by enhancements in cyberspace. These include: terrorist groups, who are employing
cyberspace to, inter alia, recruit, raise money, propagandize, educate and train, plan
operations, command and control operations; hacktivists, who are employing
cyberspace to conduct “cyber riots” (e.g., Estonia) and implement exploits in
cyberspace; transnational criminals, who pursue a variety of techniques (e.g., phishing,
denial of service attacks) to raise substantial funds (reputed to be more than the money
derived from drug trafficking); and nation states, the most advanced of whom are
employing cyberspace to enhance all dimensions of PMESII activities.
However, changes in cyberspace have given rise to unintended consequences.
Many of the entities at the “low end” of the entity spectrum (e.g., terrorists, hacktivists,
transnational criminals) are making life more dangerous for information-enabled
societies. In particular, these entities tend to be much more adaptable than nation states,
causing the latter to respond, belatedly, to the initiatives of the former. In addition,
research about selected near-peers (e.g., China, Russia) suggests that they have new
perspectives on cyberstrategy that will present information-enabled societies with new
challenges in cyberspace.
5.1.4. Institutional Factors

From an institutional perspective, issues are emerging that will affect all aspect of
cyber theory. This white paper has highlighted the challenges that exist in cyber
governance, legal issues, exchange of cyber information between governments and
industry, and the balance between national security and civil liberties.
From a theoretical perspective, one of the major challenges emerges from the
difficulty in characterizing and responding to an attack in cyberspace. As demonstrated
by recent events, it is extremely difficult to attribute an attack to an adversary that
chooses to act anonymously. In light of that ambiguity, it is difficult to formulate a
coherent response to such an attack. For example, it is still unclear how an alliance,
such as NATO, might respond in the future to a cyber attack against one or more of its
members.
5.2. Next Steps
This effort constitutes an evolving theory of cyberpower. To refine this product, it is

recommended that the following steps should be pursued.
5.2.1. Define
There is still confusion about the definitions for the key terms in a theory of
cyberpower. However, the community should find it relatively straightforward to go
from the current base to agreement on key terms (e.g., “cyberspace”). However,
additional work is still required to establish the linkage between cyber terms and the
terms associated with information operations.
5.2.2. Categorize
The “cyber pyramid” has proven to be a useful taxonomy in “binning” key concepts.
However, there is still a need to develop specific cyber frameworks and models to
explore key policy issues that confront senior decision makers.
5.2.3. Explain
It is anticipated that this evolving theory of cyberpower will be incomplete. Additional
efforts are needed to address key issues that are beyond the scope of this white paper.
In the area of cyberpower, there is a need to assess how potential changes in cyberspace
will affect political, diplomatic, and economic functionality and effectiveness. In the
area of cyberstrategy, there is a need to assess the extent to which key entities are
empowered by advances in cyberspace and cyberpower. These include individuals,
NGOs, transnational corporations, selected nation states, alliances (e.g., NATO), and
international organizations (e.g., UN). Finally, in the area of institutional factors, there
is a pressing need to assess the effect of changes in cyberspace on the balance between
civil liberties and national security. In assessing these issues it would be useful to
employ a risk management approach.
5.2.4. Connect
Currently, we have relatively little understanding about the appropriate Measures of
Merit to employ in cyber assessments nor the relationships among those measures. For
example, we do not have a clear understanding about how changes in cyberspace (e.g.,
MoPs such as bandwidth or resistance to enemy countermeasures) impacts the US’s
levers of power (i.e., P/DIME) or empowerment (i.e., PMESII). At a minimum, it is
important to develop preliminary relationships so that a decision maker can understand
the implications of how potential changes in cyberspace or institutional factors will
affect cyberpower and cyberstrategy.
5.2.5. Anticipate
As noted in this white paper, cyberspace is in the midst of explosive, non-linear
change. It is vital that more detailed technology assessments be undertaken to
anticipate and understand potential break-throughs in cyberspace (e.g., the analogue of
discovering giant magnetoresistance or fundamental changes in the architecture of the
Internet). Furthermore, efforts should be made in the development and application of

models, simulations, and analyses to assess the impact of these changes on cyberpower
and cyberstrategy. These developments in methodologies, tools, and data should
provide decision makers with the analytic support needed to explore the long-range
affect of alternative cyber options.
References
[1] Dr. Harold R. Winton, Air War College, Maxwell AFB, “An Imperfect Jewel”, presented at Institute
of National Strategic Studies (INSS) workshop on theory of warfare, NDU, Washington, DC,
September 2006.
[2] Jim Holt, “Unstrung”, The New Yorker, October 2, 2006.
[3] William Gibson, “Neuromancer”, Ace Science Fiction, 1984.
[4] Deputy Secretary of Defense Memorandum, “The Definition of Cyberspace”, May 12, 2008.
[5] “The National Military Strategy of the United States of America – A Strategy for Today, A Vision
for Tomorrow,” Joint Chiefs of Staff, 2004.
[6] Joint Publication 3-0. “Joint Operations”, Joint Staff, 17 September (incorporating change 1 13
February 2008).
[7] G.E.P. Box, “Robustness in the Strategy of Scientific Model Building, in “Robustness in Statistics”,
R. L. Launer and G. N. Wilkinson, editors, 1979, Academic Press: New York.
[8] John Markoff, “Military Supercomputer Sets Record”, New York Times, June 9, 2008).
[9] Jeremy M. Kaplan, “A New Conceptual Framework for Net-Centric, Enterprise Wide, System-of-
Systems Engineering,” CTNSP/NDU, Number 29, July 2006.
[10] Sally Adee, “37 Years of Moore’s Law”, IEEE Spectrum, May 2008.
[11] H.J. Mackinder,”The Geographical Pivot of History”, 1904.
[12] Alfred T. Mahan, “The Influence of Sea Power Upon History (1660 – 1783)”, Little, Brown and
Company, Boston, 1890.
[13] Giulio Douhet, “The Command of the Air”, translated by Dino Ferrari, Coward-McCann, 1942.
[14] Colin S. Gray and Geoffrey Sloan, “Geopolitics, Geography, and Strategy”, Routledge, 30 November
1999.
[15] David S. Alberts and Richard E. Hayes, “Power to the Edge”, Command and Control Research
Program, June 2003.
[16] David T. Signori and Stuart H. Starr, “The Mission Oriented Approach to NATO C2 Planning”,
Signal Magazine, PP 119 – 127, September 1987.
[17] Colonel Ralph Baker, “The Decisive Weapon: A Brigade Combat Team Commander’s Perspective
on Information Operations”, Military Review, May-June 2006.
[18] George Gilder, “Metcalfe’s Law and Legacy”, Forbes ASAP, 13 September 1993.
[19] Bob Briscoe, Andrew Odlyzko, Benjamin Tilly, “Metcalfe’s Law is Wrong”, IEEE Spectrum, July
2006.
[20] Daniel Gonzales, et al, “Network-Centric Operations Case Study: Air-to-Air Combat With and
Without Link 16”, RAND, National Defense Research Institute, 2005.
[21] Marc Sageman, “The Homegrown Young Radicals of Next-Gen Jihad”, Washington Post, page B-1,
June 8, 2008.
[22] Homeland Security Policy Institute, “NETworked Radicalization: A Counter-Strategy”, GWU,
Washington, DC, May 2007.
[23] Joseph S. Nye, Jr., “Understanding International Conflicts: An Introduction to Theory and History”,
New York: Pearson-Longman, 2005.
[24] Siobhan Gorman, “Electricity Grid in U.S. Penetrated by Spies”, Wall Street Journal, April 8, 2009,
page 1.
[25] Kevin Chilton and Greg Weaver, “Waging Deterrence in the Twenty-First Century”, Strategic
Studies Quarterly, Spring 2009.
[26] N. Elaine Bunn, “Can Deterrence Be Tailored?”, Strategic Forum, Institute of National Strategic
Studies, NDU, No. 225, January 2007.
[27] AFEI Conference on Cyber Deterrence, Tysons Corner, VA,Nov 1-2, 2007.
[28] John Markoff, “Vast Spy System Loots Computers in 103 Countries,” New York Times, March 29,
2009.
[29] OASD(NII)/DOD CIO Globalization Task Force, “Development of an Internet Influence/Evolution

Strategy for the Department of Defense”,October 19, 2007.
[30] M. N. Schmitt, Bellum Americanum: “The US view of Twenty-first Century war and its possible
implications for the law of Armed Conflict”, Mich. J. Int. Law 19, 4(1998), pp. 1051-1090.
[31] John Soat, “IT Confidential: Is There Anything That Can Be Done About E-mail?”, Information
Week, February 17, 2007.
[32] Audi Lagorce, “Clearwire, Sprint Nextel Scrap WiMax Network Agreement”, Market Watch,
November 9, 2007.
[33] Michael J. Riezenman, “Melding Mind and Machine”, the Institute, IEEE, June 2008.
[34] Geoffrey A. Fowler and Ben Worthen, “The Internet Industry is on a Cloud – Whatever That May
Mean”, Wall Street Journal, March 26, 2009, page 1.
[35] Convention on Cybercrime, Budapest, Hungary, November 23, 2001 (//conventions.coe.int/Treaty/
EN/Treaties/Htm/185.htm).
[36] Emad Aboelela, “Network Simulation Experiments Manual”, Morgan Kaufmann, publisher; 3rd
edition, June 2003.
[37] Ira Kohlberg, “Percolation Theory of Coupled Infrastructures”, 2007 Homeland Security
Symposium, “Cascading Infrastructure Failures: Avoidance and Response”, National Academies of
Sciences, Washington, DC, May 2007.
[38] Strategic Multi-Layer Analysis Team (Nancy Chesser, Editor), “Deterrence in the 21st Century: An
Effects-Based Approach in An Interconnected World, Volume I”, sponsored by USSTRATCOM
Global Innovation and Strategy Center, 1 October 2007.
[39] COL William Wimbish and MAJ Jeffrey Sterling, “A National Infrastructure Simulation and
Analysis Center (NISAC): Strategic Leader Education and Formulation of Critical Infrastructure
Policies”, Centre for Strategic Leadership, US Army War College, August 2003.
Glossary
Abbreviation Definition
C2 Command and control
COMPOEX Conflict Modeling, Planning & Outcomes Experimentation
CNA Computer Network Attack
CNO Computer Network Operations
CTNSP Center for Technology and National Security Policy
DAPSE Deterrence Analysis & Planning Support Environment
DARPA Defense Advanced Research Projects Agency
DIME Diplomatic, Information, Military, Economic
DoD Department of Defense
DOTMLPF Doctrine, Organization, Training, Materiel, Leadership &
Education, Personnel, Facilities
GIG Global Information Grid
HSCB Human, Social, Cultural Behavior
IAB Internet Architecture Board
IED Improvised Explosive Device
IETF Internet Engineering Task Force
IGF Internet Governance Forum
INSS Institute for National Strategic Studies
IO Information Operations
IP Internet Protocol
IRTF Internet Research Task Force
ISOC Internet Society
JMEM Joint Munitions Effectiveness Manual
JTRS Joint Tactical Radios System

JWICS Joint Worldwide Intelligence Communications System
MACS McDonnell Air Combat Simulator
MANET Mobile Ad Hoc Network
MoE Measure of Effectiveness
MoEE Measure of Entity Empowerment
MoFP Measure of Functional Performance
MoM Measure of Merit
MoP Measure of Performance
NATO North Atlantic Treaty Organization
NCO Net Centric Operations
NCW Net Centric Warfare
NDU National Defense University
NEC Net Enabled Capability
NMS-CO National Military Strategy for Cyber Operations
NRL Naval Research Laboratory
OLPC One Laptop Per Child
OODA Observe-Orient- Decide-Act
OS Operating System
OSD Office of the Secretary of Defense
P/DIME Political/ Diplomatic, Information, Military, Economic
PMESII Political, Military, Economic, Social, Information,
Infrastructure
QDR Quadrennial Defense Review
R&D Research & Development
SME Subject Matter Expert
SOA Service Oriented Architecture
SSG Strategic Studies Group
SSTR Stability, Security, Transition, Reconstruction
STRATCOM Strategic Command
TIA Total Information Awareness
TCP Transmission Control Protocol
UDP User Datagram Protocol
USG United States Government
VOIP Voice over Internet Protocol
WIMAX Worldwide Interoperability for Microwave Access
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-53
Sub Rosa Cyber War

Martin C. Libickia,1
a
RAND Corporation
Abstract. Cyberspace offers the prospect of sub rosa warfare, in which neither
side acknowledges that they are in conflict with one another or even that one side
has been attacked at all. This is possible for two reasons: first, because the battle
damage from some types of cyber attack may not be globally visible, and second
because attribution can be very difficult. The reason that both sides may keep mat-
ters sub rosa is to maintain freedom of actions, on the theory that public visibility
may complicate negotiations and lead to escalation. Nevertheless, sub rosa warfare
has it dangers, notably a lack of the kind of scrutiny that may promote actions
which cannot bear the light of day, and the overconfident assumption that no third
party is aware of what is going on between the hackers of both sides.
Keywords. Cyber space, cyber war, sub rosa
Introduction
The last twenty years have seen a burgeoning knowledge base on cyber this and cyber
that. We know a good deal more about how to get into other people’s systems – and we
know a good deal more about how to keep others out. Computer users are far more
conscious of security considerations – they have had little choice in the matter. Com-
puter security has risen in the ranks of government – and alliance – issues.
Nevertheless, it is fair to say that this growth has been concentrated at the tactical,
which is largely to say technological but also the management end. Shelves are filled
with books on how-to, but far fewer tomes explain why to. There is very little open
material on how to integrate cyber operations with kinetic operations, which is to say
how to use cyber operations to advance the ends for which kinetic operations used to be
the exclusive means. Quite possibly, there may not be much behind the green door ei-
ther, because there is very little intelligent discussion within the literature of profes-
sional military integration of hypothetical capabilities. As for strategic discussion, there
is some, but a great deal is built on the premise that cyber warfare is kinetic warfare (or
nihilistic terrorism) by other means. Well, it’s not; the two are quite different. One
might say they are as checkers and chess – which only look the same because their ter-
rain is the same and some of the pieces have the same name.
This essay expounds on one of the more interesting differences. To wit, cyber war
offers the prospect of sub rosa warfare, which is a form of combat in which the partici-
pation of both sides, or at least one side, is obscured to third parties. Sub rosa warfare
1
Senior Management Scientist, RAND Corporation. Note, the following represents the author’s view,
not those of RAND or its sponsors or clients.
54 M.C. Libicki / Sub Rosa Cyber War
has some aspects of intelligence operations, and some aspects of special operations –
although it is neither. Of note, sub rosa warfare is almost impossible to conduct with
tanks, much less nuclear weapons.
1. Embracing Ambiguity
Ambiguity, one can argue, is the essence of cyber combat and the exploitation of ambi-
guity may be central to some strategies of cyber warfare. There is a natural human ten-
dency to assume that ambiguity is an epiphenomenon, something that obscures reality,
a measurement error, as it were, and thus a temporary irritant to true understanding.
Dust the surface, and the essence of the activity shines through and we see things for
what they are. Some of this flavor comes from Clausewitz: fog and friction are what
differentiate war on paper from war in the field. Those of Platonic bent may see the
former as reality and the latter as shadows on the cave’s wall. Clausewitz warned that
one could not assume fog and friction away; they were embedded. What he did not
argue was that fog and friction were central, with all that violence being peripheral –
much less that the manipulation of fog and friction played much of a role in operational
planning.
With cyber, the opposite can be true. In some cases, ambiguity is central and dam-
age to systems is an artifact. If so, one should embrace ambiguity and not treat it as
something of an embarrassment.
1.1. Definitions
Before going further, a few boundary markers may be useful.

First, cyber war will be defined as consisting of computer network (more broadly,
systems) attack and defense. An attack succeeds when the target’s use of its own sys-
tems is hampered – either because such systems fail to work or work very efficiently
(disruption) or because systems work but produce errors or artifacts (corruption).
This definition specifically excludes computer network exploitation, which meets
neither of these criteria. It is fair to say that CNE accounts for the great preponderance
of computer network operations carried out among states and similarly serious non-
criminal organizations. Yet it is a different phenomenon. Spying is not an act of war. It
never has been, and there’s little reason to change that. Furthermore, spying is inherent-
ly sub rosa and its motives are almost self-explanatory. This is not to say that CNE
does not matter – it does – or that it is not interesting – it can be. But, it’s not the sub-
ject of this essay.
Second, retaliation will be assumed to be an option in wake of a cyber attack, but
that such retaliation will be limited to the cyber realm. This is not to say that retaliation
must always be in kind, or that both attacker and retaliator can carry on mischief in
cyber space without concerning themselves about escalation into the physical realm.
Indeed, escalation is a major motive for keeping things sub rosa. However, it is diffi-
cult to keep physical retaliation sub rosa, and considering as much takes us into a dif-
ferent conversation.
Third, the essay limits itself to state-on-state cyber war, mostly because it best
frames consideration of attack and retaliation. Non-state actors generally cannot be
deterred, or even mildly dissuaded, by putting their systems at risk – because they do
not have systems. Furthermore, although sub rosa warfare has a legitimate rationale of
M.C. Libicki / Sub Rosa Cyber War 55
sorts, the usual approach to non-state actors generally involves the application of jus-
tice, and there are serious problems with sub rosa applications of justice that transcend
cyber war issues.
1.2. Roshomon
What makes it possible to speak of sub rosa attacks is that information systems are
generally invisible. The artifacts of a system – such as a personal computer – may be
seen, but while some of what systems do is reflected back to the user, a great deal of
what they do takes place inside and is not reported. Within an organization, all of the
artifacts of computation and even most of the direct results may be hidden from the
public, and what they do is visible only to the extent that the owners wish it so. Thus,
damage to such a system is often invisible, even if some of the second-order effects
may be quite visible. The contrast with physical warfare needs no further elaboration.
Of note, therefore, is the possibility that the target, the attacker, and third parties
hold completely different perspectives on the nature of any one cyber attack.
The target includes the system operators, those they may call on for support, and
those they report to. All three of these, incidentally, may hold views of what happened
that diverge from one another’s perspectives. In general, the target ought to have some
idea that something went wrong, perhaps why, and what the consequences were. As a
general rule disruption attacks are easier to see than corruption attacks. However, the
target may not know (at least not immediately) what the attack vector did, how it ma-
naged to work past defenses, who controlled (or at least launched) the vector, and what
the purpose was.
This plausible membership of the set of people who know they have been attacked
merits further consideration. One can imagine an attack that only systems administra-
tors notice – one that, for instance, requires them to put in overtime for the purposes of
restoring a system’s prior functionality and integrity. That being so, how damaging,
which is to say, how consequential and thus how strategic can such an attack be? Thus,
one must either posit a greater affected population that, nevertheless, keeps silent (such
as the intelligence community) or a user base that senses damage but is misinformed
about why. Some potential contexts include (1) systems that were planned to go on-
stream or start new services but were prevented from doing so (a common enough phe-
nomenon without hackers), (2) systems that went down because of what was believed
to be human error, accident, or Mother Nature, or (3) systems that appear to function
normally but produce bad information that the public at large is unaware of (e.g.,
scrambled payments for medical reimbursement checks). For systems operators to keep
silent about the last one is quite risky unless they have confidence that they can correct
things later without others being the wiser.
The fourth, but partial, possibility is that the problems are blamed on hackers but
the hackers are identified as non-state actors (and thus subject to prosecution rather
than retaliation). Such an attack is only barely sub rosa, in part because many third
parties may believe otherwise (it was a state attack) and some of the implications of sub
rosa attacks, discussed below, do not apply.
The attacker includes the hackers and those they report to (assuming their reports
are honest and complete). The attacker will know the attack vector, how it evaded secu-
rity (or at least the security features they saw), and what the purpose of the attack was
(or at least one of the attacker’s bosses will know). Depending on what kind of sensors
it has emplaced in or near the attacked system, it may know something about the direct
damage, but there may be a great deal it does not know, especially with respect to pro-
grams that the target system may have to route around or make up for damage.
Third parties include the public of the target country, the public of the attacker
country, third-party states, and third-party publics. If the attack (and perhaps retalia-
tion) were really – which is to say, successfully – sub rosa they will not know much
about what, if anything is going on.
As we demonstrate, the difference in what each of these parties (broadly defined)
knows, coupled with (presumably) their reluctance to share such information, makes
the rest of the story possible.
Note that sub rosa attacks are not defined simply as those where the attacker’s
identity is unknown or uncertain – although, if the attack itself is unheard of, the identi-
ty of the attacker is moot.
Indeed, it should take little imagination to understand how much of cyber war is
subject to ambiguity: not only did something happen – but was it an accident (bad
software, human error, Mother Nature), what was the damage, what was left behind,
who did it, how they did it (including when they did it, and where was the point of
access), and, most importantly in the long run, can they do it again?
2. Basic Principles
Next, we turn to some basic principles of computer network attacks, not necessarily to
say anything original, but to emphasize a few things by way of foundation for what
comes later.
With one type of exception, the DDoS attack (more on this a little later), attacks
are enabled by vulnerabilities on the part of the target.
One can assert, for starters, that there is no forced entry in cyber space. If someone
has gotten into a system – or more particularly into the no-go area of a system – from
the outside, it is because that someone has persuaded the system to do what its opera-
tors did not really want done and what its designers believed they had built the system
to prevent. Nevertheless, in any contest between a computer’s design and use-model
(such as a user’s intuition that email is information not instructions) on the one hand
and its software code, on the other, the code always wins. Whoever gets into a system
gets into a system through paths that the software (to include protocols and firmware)
permits. The software may have flaws or may have been misconfigured (for instance,
the permissions the administrator established differ from the permissions that the ad-
ministrator thought had been established). Yet, a system is what it is, not necessarily
what it should be. Such a divergence, when it has security implications, is a vulnerabili-
ty. Whatever the methods, manual or automated, hackers’ use, an attempt to take ad-
vantage of a vulnerability to gain access to a system or to get it to accept rogue instruc-
tions is called an exploit.
A system’s integrity dictates how badly a system can be hurt by attacks in cyber
space. One might even argue that a system’s integrity is a more important determinant
of success than the quality of the adversary’s exploits—after all, no vulnerabilities, no
exploits; no exploits, no cyber attacks.
Thus, in theory, all computer mischief is ultimately the fault of the system’s owner
– if not because of misuse or misconfiguration, then because of using a system with
security bugs in the first place. In practice, all computer systems are susceptible to er-
rors. In that sense all systems are somewhat opaque, unpredictable, and thus, ambi-
guous. The divergence between design and code is a consequence of the complexity of
software systems and the potential for human error. The more complex the system –
and they do get continually more complex – the more places there are in which errors
can hide. Every information system has vulnerabilities—some more serious than others.
The software suppliers themselves find a large share of these vulnerabilities and issue
periodic patches, which users are then supposed to install – some more expeditiously
and correctly than others (notwithstanding those hackers who observe patch releases,
reverse engineer them quickly, determine the vulnerabilities the patches were supposed
to fix, develop an appropriate exploit, and use it against those slow to patch their sys-
tems). Hackers find some vulnerabilities and then spring corresponding exploits on
unsuspecting users who have otherwise done everything correctly. Literally thousands
of exploits are sitting around. Many of the more devious ones require physical access to
the target system. Most of the ones that reach the news do not work on well-patched
systems.
In a sense, cyber attacks rely on deception – persuading systems to do what their
designers do not want them to do. Fortunately, deception can be its own undoing. An
exploit, if discovered, signals to sysadmins that something is not right. If good logs are
kept, sysadmins may be able to determine where something unusual took place in the
interaction between the hacker and the system. Changes in files (data or instructions),
or the presence of unexpected files can also be telling. The process is hardly perfect; it
is possible to determine a specific vulnerability and miss the broader design flaw of
which the specific vulnerability is just an instance. Nevertheless, any one sysadmin can
take advantage of an international community of system defenders with a common in-
terest in minimizing outstanding vulnerabilities.
In contemplating cyber space, it may help to differentiate system peripheries from
the system core. Peripheries may be said to contain user equipment; that is, equipment
whose function and parameters are established by users. Peripheries, if not air gapped
or protected via consistent encryption, tend to be repeatedly vulnerable largely because
users are rarely trained in or focused on information security. User systems and privi-
leges can be taken over through password cracking, phishing, social engineering,
downloads from bad Web sites, use of corrupted media such as zip drives), etc. Sadly,
the security of the periphery as a whole is often no better than the security of the most
feckless user. The core, by contrast, is what sysadmins control—monitors, routers,
management devices, machinery (such as weapons), and databases. Sysadmins are (or
should be) trained and sensitive to security issues; they also set the terms by which us-
ers (and their systems) interact with the core. Although it is good personnel practice to
sensitize users to security issues, it is good engineering practice to assume that users
will not always be sensitive. While it is possible to protect the core from insecure users,
it is less clear whether networks can function when enough user systems are compro-
mised badly enough, even though network administration is a function of sysadmins. In
general, it is hard to compromise the core in the same precise way twice, but the peri-
phery is always at risk.
DDoS attacks are, as noted, a partial exception to the rule that a system can be at-
tacked only if it has vulnerabilities (the Mafia-Boy attack of February 2000 apparently
did take advantage of a certain class of vulnerabilities, since largely cleaned up). How-
ever, it is hard to conceive of a sub rosa DDoS attack in the sense that the public does
not notice. So, we can disregard the exception for our purposes.
sense that the public does not notice. So, we can disregard the exception for our
purposes.
2.1. The Attacker’s Motive for Going Sub Rosa
An attack can be sub rosa only if the effects are limited to entities (such as state entities
whose outputs are opaque and who believe in keeping secrets) or if the attacks could
conceivably be ascribed to something other than hacking. The target has a good deal to
say about whether an attack is sub rosa; yet, if attackers want to leave open the possi-
bility of a sub rosa attack they have to avoid having such attacks affect the broad pub-
lic but in ways that cannot be credibly ascribed to accident. They cannot take credit for
an attack, which means that it cannot be used for certain forms of coercion.
The overall motive – for both sides – for keeping matters out of the press is that
cyber warfare is a negative-sum game. Although this may be said generally true for
warfare, it may be doubly true for cyber war (CNE, importantly, aside). Simply put,
there is very little to be directly gained, which is to say, seized, in cyber war, unlike
kinetic warfare where at least one side can entertain the possibility of a smash and grab
(e.g., Kuwait’s oil fields). Cyber war cannot even disarm the other side’s cyber warfare
capabilities, and while it can disarm kinetic warfare capabilities, it can only do so for a
limited amount of time. Thus were there to be an extended cyber war, it would inevita-
bly be a contest of attrition, a test of who can, in Wellington’s terms, pound longest
before someone’s spirit gives out.
To go into particulars; an attacker may wish to limit its attacks to those that offer
the target the opportunity to keep quiet in part to forestall retaliation. The attacker be-
lieves that while the state’s elites may be able to handle things rationally – for instance,
understand when they have been back-footed and thus retreat from some position – the
same cannot be said for the target’s publics. Thus, informing such publics will put
pressure on the state to retaliate publicly when state elites may think other courses are
less costly to the state. Worse is the possibility of escalation; elites may have a consen-
sus among themselves to keep things in the cyber realm, but the public may not favor
such limits. More generally, getting one or both publics involved introduces the possi-
bility that events may spin out beyond the elites’ ability to keep things under some sort
of control. Many observers of war – for instance, of Gelb’s book, The Irony of Vietnam,
the System Worked – have concluded that state decision makers often prefer to risk
losing a war than to risk losing control over a war. Finally, if the war is controlled, it is
possible for elites of both sides to engineer a de-escalation of hostilities. All this man-
ages the risk that the attacker faces in a cyber confrontation – for both sides.
One should also note the possibility that the effects of the cyber attack can be fit
into the attacker’s narrative but only if the results of the attack can be blamed on some-
thing other than the attack. Of course, if no one notices the effects of the attack, there’s
nothing to narrate about. A sub rosa attack whose effects are felt but not explained
tends to shed focus not on the attacker but on the incompetence of the target – one una-
ble, for instance, to protect sensitive health records from being scrambled.
As noted, a high form of sub rosa warfare is to make the attack look like an acci-
dent. One should not count too highly on anything more than momentary success; in-
vestigations tend to be pretty good at getting at root phenomena.
Incidentally, for some purposes the attacker may want its identity known to its op-
posite number. A few tricks such as mailing a letter before the attack that is received
afterwards, leaving a “Kilroy was here” in the target machine, or revealing knowledge
that only a penetration could provide should suffice.
Here are a few scenarios for a sub rosa attack:
One, they can be used to put others on notice that their systems are not so reliable
that they can afford to engage in such a fight. Consider this. In step one, an attacking
state creates anomalous behavior in a key system, be it government or a government-
linked entity. The act (rather than the attacker, which is kept as ambiguous as possible)
gets the attention of the leaders of the target state, which perceives its infrastructure at
risk.
Subsequently, the system owners and their engineers claim that it was an accident
and vow that such an act will never happen again. They get large sums of money to
work hard on the problem. After this team starts to claim success, the attacker again
creates anomalous behavior, preferably to the first victim, but perhaps to another com-
parably important system. This signals that problems persist (admittedly, step two is
hard, precisely because the target state is working diligently against the possibility –
certainly on the attacked system and quite likely on similar others). This not only re-
duces the credibility of the target’s information system security, it also, and more im-
portantly, reduces the credibility of those who promised to achieve that security.
Yet the attacker does not reveal itself or what it has done. This is unnecessary and
even gets in the way. Doing so would make getting back at the attacker a more visible
centerpiece of the target’s strategy than simply misleadingly reassuring those who
know they rely on the attacked system. Indeed, the attack itself is not so much the issue
as it is to foster a general sense that the other side’s information systems are fragile and
unreliable. The attacker’s message then becomes not “Cower before us!”—which re-
quires identifying “us”—but the more impersonal, “You live in glass houses; are you
sure you want to invest so much in stones?”
Perhaps the whole point of the attack is to make the target extra wary of expanding
or opening up its networks, especially to outsiders, such as allied militaries, other gov-
ernment agencies, or support contractors. Further wariness may result from making the
attack appear to come from a trusted source. Such a strategy presumes a skewed re-
sponse from the target: not that networking should not be done naively but that net-
working is bad. It is easy to see why such a strategy can backfire and thus why cyber
strategists, thinking over an extended period, must keep second and nth-order effects in
mind.
Scenario two, cripple, test, or exercise someone else’s military. Cyber attacks on
the target’s military may be used to impede the target’s ability to respond to crises. A
large, successful attack may retard the target’s ability to wage war; if the target’s mili-
tary deployment can be delayed long enough (e.g., after everything has been decided
and after the aggressor’s forces have dug in for defense), the target’s military interven-
tion in a crisis started by the cyber attacker may be deemed pointless.
Such an attack can be a prelude to aggressive military action, or it can be in re-
sponse to fears, however ill-founded, that the target is about to start something. In the
former case, if the attack disarmed the target’s military enough to allow successful
kinetic combat, the sub rosa nature of the cyber attack may be temporary, and basically
irrelevant if the cyber attack is quickly followed by violence of an obvious sort
(“quickly” because the effects of any cyber attack are temporary and measured in hours
or days). However, if the cyber attack fails to dent the target’s military capability the
attacker may call off its dogs and has no reason to publicize what it has done. In the
latter case, cyber attack as pre-emption, the result of a successful cyber attack may be
exactly nothing – in contrast to the violence that might have happened if the target’s
systems were intact. Since the effects of the cyber attack are temporary, war may take
place anyway later – or not, if the cyber attacker (who is the presumed impending vic-
tim of the target’s military) has used the time gained to rush to the front, so to speak,
and discourage the war’s outbreak.
Complicating this logic are attacks that look like they are meant to cripple anoth-
er’s military but are not. For instance, what if the cyber attacks were meant to persuade
the target military that war was imminent, draw it to the ramparts for no reason, and
repeat the cycle often enough to exhaust or spoof the target (as Egypt did when it car-
ried out exercises in early-1973 but not attack until October of that year)? In contrast to
physical feints, however, cyber feints may be poor strategy. By hardening the target’s
systems, every attack makes a subsequent attack more difficult. The choice of targets, if
not masked by noise, may also suggest what the attacker finds important to disrupt and
thus hints at how the cyber attacker would fight if war turned physical.
Attacks may be launched on military systems to see how well their operators react,
in preparation for some later, larger attack. Can enemy sysadmins determine what hap-
pened and why? What workarounds do they use? Will corruption be detected? If the
target knows it has been so tested, should it retaliate? Conversely, attacks may well
reveal a great deal about the attacker and what it knows about the target’s vulnerabili-
ties.
There are solid grounds for believing that attacks on military can retain their sub
rosa character. The attacker has the usual motives for keeping quiet, with the possible
exception that it may wish to whisper about the attack to the target’s allies so as to re-
duce their faith in the target’s military. For the target, on its part, to reveal that it was
attacked – and successfully so – is apt to reduce rather than increase confidence in its
military capabilities. The latter may not have much of a choice if the damage is so
widespread that a universe of witnesses defeats all thoughts of keeping them silent. The
target may also broadcast the attack for purposes of supporting a “hate the enemy”
campaign, regardless.
Cyber attacks that cripple intelligence assets do not have to lead to war. They may
be justified if they blind the target’s systems long enough for the attacker to carry out
operations (e.g., moving missile parts) safe from prying eyes. Perhaps needless to add,
intelligence assets are extremely hard targets for cyber war.
Coercion – especially against democratic states – normally requires the damage to
be publicly visible and clearly associated with the coercer and its cause. Adversary
actions need not affect the public, though, if there are other ways to compel govern-
ments to accede to demands. Indeed, the opposite may be true: the less the public
knows, the easier it may be to garner concessions, especially invisible ones.
The case for sub rosa cyber war for the purposes of coercion rests on the belief
that publicly visible attacks could lead to more popular pressure on the state to stand
firm than to concede. The attacker counts on the possibility that the target’s leaders are
less afraid to make concessions whose true rationale can be hidden than to be blamed
when, say, the economy hits an air pocket. As long as the new policy (which contains
concessions) does not appear unwise per se or does not contradict earlier policies too
much, the target’s leadership need merely hide the fact that their policy choices were
driven by fear. Keeping mum has other advantages for the target. Reducing the public
itch for revenge (or their desire to demonstrate resolve) may facilitate negotiations or
mutual de-escalation. Obscuring the fact or at least the damage from the attack may
also mask the state’s vulnerabilities from the eyes of third parties (presumably, the at-
tacker will have a better sense of which vulnerabilities it had, in fact, exploited).
One ought not forget in all this that the sub rosa strategy has a serious Achilles
heel from the cyber attacker’s point of view. It assumes or, more to the point, requires
that the target reacts as expected and maintains its silence. This requires that the cyber
attacker have sufficient insight into the target to operate below the threshold past which
it decides to mobilize against the cyber attacker – an act that generally requires the tar-
get being open about the attack and its consequences. The larger the cyber attacker’s
gain vis-à-vis the target, the less likely the target is to restrict its own activities. In ef-
fect, the attacker’s strategy is hostage to the target’s behavior, the basis for which we
now turn.
2.2. Should the Target Reveal the Cyber Attack?
The likelihood that any attack is visible is the likelihood that the effects of an attack are
visible multiplied by the likelihood that these effects will be publicly ascribed to a cy-
ber attack (rather than to error, accident, or bad design). Both parts of the equation are
anything but given. CNE is rarely apparent until an investigation reveals it. Corruption
may go unnoticed until it reveals itself as a discrepancy between what a system is doing
and what it should be doing. Sometimes even disruption may go unnoticed; for exam-
ple, if a sensor is silent, is it silent because it has nothing to report, or has someone
tampered with its reporting channel? If it is not people but machines or other processes
which consume certain services, their loss may be noticed only when the processes
they feed behave incorrectly.
Normally, full disclosure is the best policy. It is too easy for governments to be-
lieve they can control information much better than they actually succeed in doing –
witness Chernobyl. Post hoc revelation eats at government credibility—not to mention
competence, if playing catch-up with events makes the government look bad. Scream-
ing helps mobilize the citizenry to support the government and (less cynically) pay at-
tention to information security. It raises the seriousness level of the whole cyber war
contest and thus gives the government more scope for implementing domestic security
measures that the citizenry would otherwise object to. If the fact of the damage is evi-
dent, but not the cause, revealing the cause may enhance the credibility of infrastruc-
ture owners by switching attention from their own fecklessness as sysadmins to factors
(portrayed as) outside their control. Revelation is necessary if the target state is going
to respond visibly, either with retaliation or without (using legal, diplomatic, or eco-
nomic measures, for example). Going public provides an opportunity to be clear about
the aims of the response; it also subjects them to the test of knowing whether it can
bear scrutiny. Incidentally, revelation may also be necessary for sub rosa retaliation:
just because the retaliator did not want to make a fuss about how it hit back does not
mean that the attacker (as target of retaliation) will do likewise.
Yet silence may still be golden. Revelation may expose the fecklessness of the tar-
get’s system security, reducing the public confidence in it and making it a target for
repeat attacks (a case for discretion comes from the public’s tendency to overestimate
the risks of cyber insecurity; there is considerable agreement that the public is wildly
inconsistent in how it reacts to low-probability, high-impact risks). Evidence to support
the attack claim may reveal sensitive information about system security.
2.3. Should Cyber Retaliation Be Obvious?
In cyber space, the target can hit back against the attacker, and no one (aside from the
security establishments on either side) need be the wiser. This sort of sub rosa retalia-
tion tends to make more sense if the attack is not public or if public attribution is not
viable. In the latter case, the evidence behind attribution may be of the sort that is not
easily released or not easily argued if released. Sub rosa retaliation avoids having to
make the choice of what to reveal. This is no small matter. Reveal one’s forensics and
one has given all attackers a clue about what to avoid leaving behind the next time.
Information about sources and methods is among the most closely-guarded secrets of
the intelligence community. Furthermore, the attacker, as the victim of retaliation,
could be under subsequent public pressure to counter-retaliate. If the effects of retalia-
tion were not obvious, the attacker could therefore conclude that letting things drop
after the retaliation is wiser than carrying on.
States that would employ sub rosa retaliation have to manage the expectations of
those in the know who are looking for revenge. Retaliation could still convey the tar-
get’s displeasure over the attacker’s leadership and could change the latter’s calculus to
discourage further attacks.
Sub rosa retaliation, however, may be too seductive, particularly if the retaliator
feels no need to convince the attacker of its guilt—after all, the attacker knows that it
struck first, right? One danger is that, if the intelligence or law enforcement agency
does not need to worry about defending its attribution to others, its case to national
command authorities (that is, those who control the retaliation capability) may go un-
challenged. The agency may thus claim its attribution is correct when the evidence
suggests a higher degree of skepticism is warranted. Furthermore, a decision to retaliate
sub rosa – like the decision to attack sub rosa – takes certain targets off the list (e.g.,
power plants) or at least demands they be hit in ways that do not look like a hit (which
then fails to communicate displeasure reliably). The remaining targets may be those
thought to be important to the other side’s intelligence and law enforcement communi-
ties but do not directly affect the public at large. Finally, the entire strategy rests on the
attacker’s willingness not to make a fuss. Again, but in reverse this time: the wisdom of
the strategy is hostage to the discretion of the state that (supposedly) engineered the
attack in the first place.
2.4. Sub Rosa Retaliation against a Sub Rosa Attack Has One Big Advantage
To wit, the requirements for attribution are not nearly so great. One does not even need
that much confidence in the quality of attribution. So, in hitting back, one may consider
two possibilities. One; it was the attacker that suffered retaliation. Two; it was an inno-
cent third party.
Take the first case. The attacker, knowing that it started things, will have a fairly
good idea of why it was hit and take the message (subject to all the other caveats). If
retaliation is to be reliably read as retaliation by the attacker, the “accident” would have
to occur rather quickly after the original attack. Thus, the capacity to retaliate has to be
maintained at a fairly high degree of readiness (that is, one must ascertain that the vul-
nerabilities still exist and that the victim’s reaction will be roughly as predicted). Fur-
thermore, the normal deliberation that might take place after an attack to increase the
odds that the retaliation was well-directed would have to be short-circuited. All the
previous caveats about the difference between what you think others do not know and
what they actually do not know also apply.
In the second case, the innocent third party, unaware of what may have motivated
an unprovoked attack (which the retaliation may look like to the victim) can only trot
out its usual suspects and look for forensic evidence. As noted, this requires the origi-
nal attack be unknown to any but the attacker and the target.
Managing the consequences of any venture that assumes ignorance among others
is always contingent on third parties not spilling the beans. For instance, if retaliation
against a third-party state is discovered by the attacking state, the attacker now has a
very valuable piece of information – who attacked the third-party state. If the attacking
state can figure out how to profit from implicating the retaliating state, it may well do
so. Telling the third-party state that it started things may not be the smartest move, but
it may be able to downplay its own role to suggest the retaliator over-reacted and was
stupid about things to boot. It may maintain its innocence but circulate hints that make
it easier for the innocent victim to identify the attacker (finding something is a lot easi-
er when you know exactly what you are looking for). Or, the attacking state may
blackmail the retaliator lest its actions be revealed to the innocent victim. The assump-
tion that no one in the third-party state knows about the original attack may be in error;
it is not unknown for two states with little in common but their dislike of the United
States to swap intelligence (Iraq and Serbia, for instance, traded information on how to
defeat U.S. aircraft and avoid anti-radiation missiles). More generally, the original at-
tack may not be so secret prior to the attack or its existence may be revealed after the
fact. Such revelation may be deliberate (perhaps someone here in the know is bothered
by the retaliation or the possibility that it was misdirected), or simply reflect the univer-
sal difficulty of hiding secrets. Finally, the retaliator may have overstated its ability to
keep itself anonymous. The third party does not have to know who did it, but it may
have serious enough suspicions to affects its relationship with the retaliator – and if it
did not know why the retaliator acted as it did, it may be angrier than if it understood
that retaliator’s motivation.
Again, perhaps sub rosa cyber war may be too clever by half – and one does not
gain points for upholding the rule of law in cyber space by being sneaky.
What about being even cleverer and making retaliation look like an accident? The
last technique is a variant of the first. Not only is the retaliation anonymous but it ap-
pears to be an accident. It is two steps rather than one step removed from something
that the innocent third-party victim may find actionable. Again, the true attacker will
presumably suspect that the accident was too closely timed to the original attack to be
an accident, while the innocent victim of misguided regulation will have even less indi-
cation of what happened much less why. Indeed it would be most cool if the reprisal
could be made to look like something caused by the original attack going haywire – all
the dissuasive impact, and none of the risk.
It is unclear how to make an attack look like an accident in the first place. True,
many attacks are initially hard to distinguish from accidents – which argues against
hasty reactions all around. But there are techniques that can distinguish the two. If the
problem is faulty software (such as the DSC bug that crippled phone service in the
1990s) then the fault can often be replicated by simulating the conditions at the time of
failure. Human error can often be detected in various process logs. The greater the pain,
the greater are the resources likely to be devoted to its elucidation. Thus, safeguards
against the victim’s (whether the original attacker or the unfortunate third party) detect-
ing that it has been attacked may be temporary.
Finally, a state that wishes to establish principles – such as, do not hack – and then
enforces them surreptitiously communicates either that it lacks sufficient faith in such
principles or the strength to maintain and defend them openly.
Subtlety, nay sneakiness, in retaliating against a cyber attack absent strong attribu-
tion is normally difficult, but the exigencies of cyber space – the high level of ambigui-
ty everywhere in the medium – only make things harder. Thus, while there are some
notional ways to ways to work around the attribution problem they require a great deal
of certainty about matters (the effect of cyber attacks, or the perception of attackers and
third parties) that stand in stark contrast to the uncertainty about who did it. This leaves
us with approaches that our British friends might call frightfully clever, with the em-
phasis on “frightful.”
2.5. Ending Sub Rosa Warfare
How does one end a war that one does not admit one is fighting? In general, wars can
end in one of four ways: through the destruction of one or both parties, through a for-
mal peace agreement, through an informal tacit peace agreement, or as a series of bila-
teral decisions not to attack.
Cyber war generally lacks the power to destroy one or more parties to a conflict,
and all the more so when the warfare is sub rosa – which not only takes certain types of
attacks off the table (notably those that put pressure on populations), but also lies below
the level where either side has reason to escalate into at least explicit warfare.
A formal peace agreement that pledges each side to halt an activity appears incon-
ceivable if neither party admits to being the victim, much less the perpetrator of the acts
in question. Yet, the transition from sub rosa to explicit cyber war is easy to make.
Third parties may discover as much and make their findings public. Each side may also
discover reasons for changing its mind and announcing as much.
An informal peace agreement requires that each side of a fight that is not public is
nevertheless willing to discuss such secret maneuverings with its counterpart. At a min-
imum this requires some confidence on each party’s part that it is not kidding the other.
Both formal and informal peace agreements in cyber space, however, can be prob-
lematic to enforce, or even state the terms of. Monitoring peace pacts in cyber space
poses challenges not found in physical space. If either side still believes it can, if unpu-
nished, reap unilateral advantages from new attacks, then attribution and damage as-
sessment will likely remain as difficult afterward as they were beforehand (if attacks
are extended to include CNE, the odds that one or another side finds attacks useful only
go up). Each could cheat by shifting from visible disruption attacks to more-subtle
corruption attacks.
Unfortunately, tacit de-escalation presents many of the same validation problems
as negotiations – only made worse by the fact that there would only be a rough consen-
sus rather than an explicit statement of what actions were and were not considered a
violation. How could one tell that the other side is even cooperating, without clarity on
what constituted cooperation?
In physical wars, peace pacts are often followed by unilateral disarmament (after
World War I, for instance, Germany’s army was limited to 100,000) or multilateral
disarmament (for example, the Washington Naval Treaty). But disarmament in cyber
space is virtually meaningless because cyber war is less about arms (exploits) than
about vulnerabilities. So, disarmament cannot bulwark a peace agreement that applies
to cyber space.
Mutual transparency may help keep the peace (in much the same way that formerly
warring sides exchanged hostages), but no state (not even a friendly one) exposes the
secrets of its security architecture to another. Besides, if the war is still sub rosa both
sides have amply demonstrated the virtue of transparency. If it did, the transparency
would have to be bilateral rather than public, lest mischievous third parties profit from
the new-found knowledge. Even then, each side could attack the other from third par-
ties outside the transparency agreement.
Thus, the least problematic outcome is for neither side to find any especial reason
to commit serious resources to breaking the systems of the other. This may ensue be-
cause the broader ends that led at least one of them into cyber war in the first place
have been met or because further cyber war will get no party closer to meeting them
than the last spate of cyber war did.
The part of the equation in which one side decides that the effort no longer pays is
not strategically problematic because it does not require the other side to recognize that
anything has changed. But it is hard to believe that the party that quit making the effort
would not hope to see some rewards for its restraint. As long as the one side had not
made either explicit (that is, negotiated) or implicit commitments to restraint, the other
side would not be able to hold up some future system malfunction as evidence that it
had been lied to or cheated. Furthermore, if the other side still found advantage in com-
puter attacks – or if it was engaged in other forms of hostilities – it may have no motive
to acknowledge such restraint. But if the other side also finds that the advantages of
hacking have waned or that they are trumped by the rewards of friendly engagement, it
too might work itself into a modus vivendi.
3. Conclusions
Cyber space is a medium in which the absence (or, more specifically, unimportance) of
physical artifacts permits a form of warfare that is generally unavailable in other media.
The closest analog to sub rosa warfare would be a campaign of espionage, but even
there, the potential exposure of the saboteurs (whilst hackers can be sheltered by the
attacking country or in the anonymity of the Internet) makes it hard to keep matters
quiet for terribly long. That such sub rosa warfare is possible, however, makes it nei-
ther probable nor particularly wise. Paradoxically, maintaining sub rosa warfare re-
quires the tacit assent of the other side, and is therefore quite fragile. More practically,
the very shadowy nature of the whole enterprise (coupled with the difficulty of getting
policymakers to understand the requisite ins and outs of cyber war in general) creates
enormous temptation to take risks without adequate political consideration of their cost.
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-66
Warfare and the Continuum of Cyber

Risks: A Policy Perspective
Andrew CUTTS
U.S. Department of Homeland Security
Abstract. At the highest levels of national government, two of the most important
decisions to get right are properly prioritizing among competing missions, and
balancing between short-term and long-term objectives. The most consequential
and highest risk threat is attack by one or more nation-states intent on projecting
power, and who are willing to damage or destroy critical information infrastructure
by cyber means in order to achieve this objective. Threat actors falling into this
category have the necessary time, resources, sophistication, and access to do so.
This category certainly includes cyber warfare. Today, nation-states are beginning
to understand in concrete terms the potential benefits and costs of cyber attacks
used as a means of projecting national power. It may not take a great deal of a
nation’s cyber resources, planning time, or technical access to achieve limited
national objectives.
In the U.S., cyber defense of critical infrastructures is largely a homeland security
mission. It may be that defense always lags the most potent offense. But the goal is
an effective defense, not a perfect one. To get ahead of the most serious national
cybersecurity risks, including that of cyber warfare, a country’s cybersecurity
leadership must seek an appropriate balance of resources, energy, and focus
between those threats that are most frequent and those that are most consequential.
The historical bias in dealing with cyber risk has been to look at it through the lens
of commerce, not national security – and to reinforce the emphasis on short-term
thinking rather than long-term strategy. One way to overcome this bias is simply to
emphasize efforts that mitigate the most consequential risks. A nation’s cyber
leadership could decide, for example, that it should apply significant early
resources to mitigating the national security risk associated with defending critical
infrastructure against nation-state threats.
Keywords. Cyber warfare; critical information infrastructure; cyber risk;

cybersecurity policy; cybersecurity; homeland security; cyber attacks
Author’s note
This paper offers a framework for thinking about and debating vital national
cybersecurity policy issues. It makes no attempt to settle those issues. Its primary
purpose is diagnostic. To the extent it offers prescriptions, it does so only to forward
thoughtful debate and discussion. It does not reflect a settled position of the
Department of Homeland Security or of the U.S. Government.
A. Cutts / Warfare and the Continuum of Cyber Risks: A Policy Perspective 67
Introduction
The cyber attacks against Estonian networks in 2007 were a wake-up call for
information-based societies in general, and for the North Atlantic Treaty Organization
in particular. Those attacks demonstrated that protecting classified networks and
defense-related communications, while very important, is insufficient for an
information based nation-state.[1] They forecast the risk that critical information
infrastructure owned and operated by the private sector, including that which supports
energy, transportation, banking, communications and the media, could in the future be
the target of cyber attacks by a strategic opponent. The defense and security of these
networks is in the national and public interest. A country’s national security and
economic well-being are at stake.
NATO as an organization can do much to establish and enhance a common cyber
defense among its members. Yet it is largely up to each nation to protect its own
respective networks and infrastructure from cyber attack. In this respect, each of our
countries is very much in the same boat. To a greater or lesser extent we each face
similar challenges in protecting nationally vital information infrastructure.
1. A Simple Conceptual Framework
At the highest levels of national government, two of the most important decisions to get
right are properly prioritizing among competing missions, and balancing between
short-term and long-term objectives. This is true in monetary policy. It is true in energy
and environmental policy. It is vital to military success. And it is no less vital to
national cybersecurity efforts.
Dr. Paul Bracken, a professor at Yale University, once wrote of the need to “model
simple, and think complex.”[2] Simple models help us to think about the complexity of
what they reveal. A simple model1 the author has found useful for thinking about and
communicating the need to balance between competing national cybersecurity
priorities is the cyber risk continuum in Figure 1 below2.
The graphic depicts a range of cyber threats, increasing in both potential
consequence and risk from right to left3. In this case, consequence can be thought of as
the potential for harm to a nation’s security or economic well-being. These threats are
not unique to the U.S. but rather are faced to some extent by any information-based
society. The graphic conveys that the threats of highest consequence, and their
associated risks, are also likely to occur with the least frequency.
On the far right of the continuum are nuisance threats including “script-kiddies”
and hackers. Next in increasing consequence is cyber crime. Criminals, of course, are
1
The term “model” is used to convey that this is not a quantitatively-driven plot of data.
2
The model is not backed by a data-driven assessment of risk. It is based on anecdotal evidence.
3
This paper frequently uses the term “cyber risk”, which differs from a “cyber threat”. Risks are
combinations of threats, vulnerabilities and consequences. A discussion of specific vulnerabilities is outside
the scope of this paper, which focuses on two of the three factors in determining overall risk – namely
“threats” and levels of “consequence”. The paper assumes vulnerabilities exist, and that threat actors differ
in their capability and motivation to exploit these vulnerabilities in ways that might harm a country’s
economic or national security.
68 A. Cutts / Warfare and the Continuum of Cyber Risks: A Policy Perspective
National Cyber Risk Continuum

Severe (logarithmic scale)
Project power / damage or destroy

Nation-state / unlimited resources
Consequence
Nation-state / terrorist
Historical progression
of risk
limited resources
Project power
Nation-state /
Very Low
Criminal
Hackers
gang
Steal
Very Low High

Frequency
Figure 1.
motivated to make money and are willing to break national and international law to do
so. Cyber criminal gangs are increasingly sophisticated and the economic cost
of their illegal activity – most notably in the banking and finance sector – has
greatly increased over the last five years.
Nation-states that are capable of and motivated to steal intellectual property or
state secrets (espionage) pose the next threat on the chart. This is due largely to the
potential resources, sophistication, and motivation they apply to achieve their
objectives.
The next most consequential threat on this continuum is that posed by either
nation-states or non-state actors, including terrorist organizations, which might be
motivated to project power through cyber attacks on critical infrastructure. This
category includes threat actors who have limited resources, time, and access to
accomplish their objectives.
The most consequential threat is attack by one or more nation-states intent on
projecting power, and who are willing to damage or destroy critical information
infrastructure by cyber means in order to achieve this objective. Threat actors falling
into this category have the necessary time, resources, sophistication, and access to do
so. This category certainly includes cyber warfare.
It is important for national decision-makers to understand that cyber warfare exists

on a continuum of risk. It is not going too far out on a limb to say that risks have tended
to increase over time. Effective mitigation of lesser threats does not necessarily mean
that more consequential threats, including cyber war, are also mitigated. It is obvious,
but should be stated nonetheless, that a nation’s tolerance for cyber risk relates directly
to its economic and security dependence on information infrastructure.
2. Limitations of the Model
It has been wisely said that “all models are wrong, but some are useful.”[3] In that
spirit, the author acknowledges that the continuum is overly simplified. It shows only a
few threat categories from among many combinations of threat actors and capabilities.
It treats these as distinct from one another, whereas in reality nation-states, terrorist
organizations, hackers, and cyber criminals might use one another, perhaps
unwittingly, to achieve their respective objectives. And of course it is often impossible
to tell one threat actor from another in cyberspace, given the inherent difficulty of
attributing attacks to their ultimate source. Still it is useful as a construct for thinking
about the problem, and for conveying some of its strategic implications to senior
decision-makers.
First, it conveys the simple progression of the cyber threat – a key component of
risk. Early on in the development of cyber infrastructure, we were only exposed to risk
from script kiddies and hackers. As more businesses connected to the Internet,
sophisticated criminal gangs emerged. Within the last few years nation states are
reported to have stolen massive amounts of data from defended networks. In other
words, the progression of risk has steadily moved to the left. The author believes the
underlying reason for this is economic. At each step the benefits of cyber attacks to
those who conduct them have outweighed the costs they incur.
This was acceptable when the potential consequences were low. But they are
increasing. The current state of affairs is clearly unacceptable. In the U.S. it has been
called a national security crisis.[4] Today, nation-states are beginning to understand in
concrete terms the potential benefits and costs of cyber attacks used as a means of
projecting national power. It may not take a great deal of a nation’s cyber resources,
planning time, or technical access to achieve limited national objectives.
As every businessman knows, past performance is not necessarily an indication of
future activity. One cannot predict that just because the progression of risk has moved
steadily and swiftly up the continuum, it will continue to do so. But an obvious
question for national policy makers is this: “What set of factors might stop this
progression?” A corollary question: “Should we assume the progression will continue
unless the economics of the problem changes – unless costs to potential attackers can
be introduced to clearly change their cost/benefit calculation?”
One way to change, or at least slow, the vector of risk is to raise the cost of attack
by enabling a better defense. In the U.S., cyber defense of critical infrastructures is
largely a homeland security mission. It may be that defense always lags the most potent
offense. But the goal is an effective defense, not a perfect one. The exact nature of that
defense will vary from country to country, sector to sector, network to network, and
threat to threat. It will depend on strategic objectives, an assessment of strengths and
weaknesses, available resources, national capacity for research and development, and
tolerance for risk, among other factors. To be effective, a national cyber defensive
capability, commensurate to the level of risk, should exist before a country experiences
the high consequence threat on the left hand side of this continuum.
3. An Apt Metaphor
Several years ago a United States Senator, The Honorable Robert F. Bennett, gave a
keynote speech at a conference on cyber conflict. He used a sports metaphor to convey
a key point. Referring to Wayne Gretzky the famous hockey player, Senator Bennett
alluded to the fact that Gretzky had such a sense for the flow of the game, he could
anticipate ahead of other players where the puck would be – and he skated to that
position on the ice. He arrived before the puck, and was then in position to help the
team score.
The author believes the Senator’s point was this: an information-based society like
the U.S. cannot protect its information infrastructure from the worst cyber risks unless
it makes a concentrated effort to get ahead of the threat. This point is vitally important.
Yes, our countries must address the threats we face today; we cannot neglect them. But
for some countries it could take years, perhaps decades, to build an effective defense
against the most consequential risks on the continuum. It is one thing to protect
government networks. It is entirely another to protect non-government networks
against nation-state cyber threats. Building a national capacity to do so will not happen
overnight. And that raises another vital question for any national policy maker: “How
long do we have before the most consequential threats might materialize?” Whatever a
country believes that timeframe to be, if it has no effective defense in place before then,
it assumes a very great risk indeed.
4. A Policy Imperative
The cyber threat never stops. Our respective operational echelons and cyber defenders
have no time to come down from the ramparts. Their typical day is filled with efforts to
mitigate current and near-horizon threats. But over-the-horizon risks will not disappear.
Operational cadres may not have the time or present capability to deal with them, but
these risks deserve more than a fleeting glance when operations allow. A country
whose tolerance for cyber risk is low should devote the resources necessary to
understand the most consequential threats and address the risks they pose.
Nobel Prize-winning economist Herbert Simon once said, “Short term thinking
drives out long term strategy, every time.”[2] This insight is the economic corollary to
the Gretzky metaphor. It certainly rings true in the field of cybersecurity.
The national cyber risks faced by an information-based society are great. They
may seem far in the future; but the most consequential risks must be mitigated today
with action that is direct and decisive, not oblique or incremental, regardless of their
frequency. Proactive steps to mitigate over-the-horizon risks will be much less costly to
commerce and national security than allowing these threats to materialize. Recent
experience in other policy domains, including finance and hurricane preparedness, has
proven this point.
To get ahead of the most serious national cybersecurity risks, including that of
cyber warfare, a country’s cybersecurity leadership must seek an appropriate balance of
resources, energy, and focus between those threats that are most frequent and those that
are most consequential. Creating the conditions in government where infrequent threats
can be understood and addressed is easier said than done.
In each of our countries, the organizations that have defensive cybersecurity
responsibilities perform one or more of three different missions. They fight cyber
crime. They defend government networks, including those that are used by civilian
agencies, the military, and the intelligence community. And in some cases they must
help protect non-government networks that qualify as critical national infrastructure.
As we all know, these are mostly owned and operated by the private sector. They
include the data, hardware, software, and control systems that undergird our financial
markets, the generation and distribution of electricity, modes of mass transportation,
and our vital telecommunications. They support thousands or millions of competitive
business models - each one unique; and they are operated mostly with economics in
mind.
This last mission raises two additional policy questions for many of us. The first is
this: “Against which cyber threats on the continuum should our governments be held
responsible for protecting the private sector?” At every point on the continuum,
commerce is vital. So are civil liberties. Clearly, the bias at the lower end of the risk
spectrum should be weighted toward private enterprises taking the lead for managing
these risks as they relate to individual business models. Equally clearly, no private
enterprise – no matter how well capitalized – can bear the cost of defending itself
against destructive nation-state attacks. In this case the opposite ends of this continuum
are a bit like the opposite poles of a political spectrum; it is fairly easy to see what
exists at either end, but it is much harder to characterize the middle.
5. Drawing the Line between Security and Defense
This leads to the second question: Where on this continuum should a country’s
leadership draw the line between security and defense? Where does one stop and the
other start? A country cannot debate this question forever. Leaving it unanswered leads
to a situation in which no one – not the defense community, not the security
community, nor the private sector, is clear about responsibilities. Lack of mission
clarity leads to lack of authority, resources, and capabilities. And that comes at the
expense of neglecting the high end of the consequence spectrum.
Figure 2 depicts one way of thinking about the difference between cyber “security”
and cyber “defense” – at least for a western democracy such as the U.S. It shows two
parallel lines - both of which are drawn rather subjectively 4. The area below the grey
4
Exactly where the security and defense boundaries should be drawn in relation to the continuum of threats
is worth careful policy debate and consideration. In this case they are drawn for illustrative and discussion
purposes only; in reality they could be higher or lower. Moreover, it may well be that the higher end of the
“security” boundary is not static across all threats, but rather it increases in stair-step fashion as the threat
increases. By this is meant that the private sector might be enabled to participate at higher and higher levels
of capability in their own (and the national) defense as threats and risks escalate.
line is labeled “security”. In this case, the term “security” indicates that the clear bias
should be toward expecting private enterprises to bear primary responsibility for
managing risks in this range. Naturally they would do this primarily out of fiduciary
responsibility to their stockholders – but also in some cases as part of a regulatory
framework.
This does not mean that the private sector should be solely responsible for
mitigating risks associated with threats at the lower end of the spectrum. Cyber crime,
for example, is an area where the private sector must work with law enforcement to
address the threat adequately. Many national Computer Emergency Readiness Teams
(CERTS) also provide incident services to the private sector that help them mitigate
risks even from the lower tier of cybersecurity threats.
It does mean, however, that both government and industry should agree that the
primary metric through which the risks associated with these threats are cooperatively
managed is that of maintaining competitive business models. Mitigation efforts must
sustain profitability for individual businesses, value chains, and complete sectors of
business activity.
Figure 2.
Note that on this graphic the “security” domain extends across the entire spectrum
from left to right. One important conclusion is that private enterprises – at least those
that operate critical infrastructure - must have some responsibility, presumably backed
by demonstrated capability – to assist in mitigating all categories of cyber threat - even
those that contribute to the most consequential national cyber risks.
The extent of that responsibility is a question of strategy. At its most basic, the
central question is this: should privately-owned critical information infrastructure be
defended against the most consequential threats from “within the enterprise” – meaning
by experts who operate and know the importance of each byte on those networks on a
daily basis? Or should they be defended by government cybersecurity experts, using
nationally developed capabilities from outside the normal perimeters of most business
networks? Both have advantages. Both also present very difficult challenges. A
balanced strategy might include both, but the question then becomes: what is the
appropriate balance between the two? Answering this question is fundamental to
national cybersecurity efforts.
The area above the blue line is labeled “defense”. In this case the term conveys
that risks resulting from these threats are systemic and could be nationally debilitating.
Government should be primarily accountable for addressing this tier of threat, even
though operational responsibilities must inevitably be shared between government and
the private sector. Indeed, the threats at this end of the spectrum are so potentially
severe that their mitigation should be thought of as a mission to be performed rather
than as something to be managed.
These threats are simply beyond the scope of private sector capability to
adequately address. Both government and industry should agree that the primary lens
through which these high-tier threats are addressed is one of national security. If
national security is at risk, commerce is also at risk. There should be no question about
priority in this domain. National security must take priority over commercial interests
for those who are assigned responsibility to manage the highest consequence threats,
whether that assignment is given to the defense community or the homeland security
enterprise.
6. The Sticky Middle Ground
In between is a shaded area in which the delineation between security and defense is
blurred. It is in this area where roles and responsibilities between government and
industry are most difficult to define. Addressing threats that fall into this domain offer
the most difficult decisions. This is true for two reasons.
First is the strong potential for conflicts of perspective and for competing interests.
The private sector must compete; fiduciary responsibilities require businesses to
maintain their focus on the bottom line. In most cases, their risk horizons are invariably
short. On the other hand government must support and enable commerce, but not at the
expense of providing for the common defense – a constitutional requirement in the U.S.
It must take necessary steps to manage long-term risks. Inevitably, tension exists
between these differing perspectives and responsibilities.
Second is the extent to which mitigation decisions for risks in this middle tier
involve unknowns. The complexity of cyberspace; its tendency to create unforeseen
interdependencies; the way it immediately transmits and links impacts of decisions

made remotely across great distances and geographic, political, and organizational
boundaries; the potential for small, hidden vulnerabilities to result in highly leveraged
consequences; and its newness as a domain for conflict, all greatly increase the fog of
risk management and crisis decision-making. This is especially true in the domain of
risks that occupy the middle of the consequence spectrum.
This leads to another important insight. It is incredibly easy to get bogged down in
mitigating mid-tier risks. Recall that the progression of threat – and by extension the
progression of risk- reached this middle tier by starting at the bottom end of the
consequence spectrum. The historical bias in dealing with cyber risk has been to look at
it through the lens of commerce, not national security – and to reinforce the emphasis
on short-term thinking rather than long-term strategy.
These factors, together with the tendency for conflicting perspectives and difficult
decision-making against this middle tier of threats, create an operational environment
in which the struggle to devote any meaningful time and effort toward getting ahead of
the most consequential threats is a real challenge.
One way to meet this challenge is simply to emphasize efforts that mitigate the
most consequential risks. A nation’s cyber leadership could decide, for example, that it
should apply significant early resources to the left end of the continuum – to mitigating
the national security risk associated with defending critical infrastructure against
nation-state threats. Over time it could capitalize on these resources by applying them
against lesser risks. In this way it could ensure that it does not grind to an operational
halt short of accomplishing its highest strategic priorities. It also could gain the most
value from its resources.
Other ways to emphasize the most consequential threats and risks: (1) develop a
long-term war-gaming practice that continually refines the policy, legal, economic,
operational, and technical issues associated with the high end of the continuum; (2)
ensure planning scenarios for exercises and war-games focus on these threats; and (3)
require that national and sector risk assessments cover the entire risk continuum.
A logical step for any country’s cyber leadership is to undertake a continual effort
to assess risk across this spectrum. Part of this effort should include identifying the
subset of discrete vulnerabilities in critical information infrastructure, which if
exploited would have the most debilitating consequences to national or economic
security. Developing an appropriate strategy for mitigating each of these discrete risks
should be a joint effort between government and the private sector. But government
owns the responsibility, and should have the authority, to say when or whether the most
severe risks have been acceptably mitigated.
7. Capability v. Mission
Of course, delineating risk responsibilities between government and the private sector
is only part of the solution. In large bureaucracies such as in the U.S. the imperative to
clearly define defense from security exists for another reason, and that is the need to
clarify missions between government agencies, and assure each mission is supported
by adequate capability.
If one’s mission responsibilities are unclear, it is impossible for one to know if his
capabilities are sufficient.
Figure 3 shows the threat continuum overlaid against an assessment of capability

vs. mission. It supposes the existence of an organization with a given risk mitigation
capability. If this organization is assigned a mission of defending critical information
infrastructure against criminal threats and theft of intellectual property, Mission (A),
then its capability is sufficient. If however, its mission includes defending the same
infrastructure against destructive cyber attacks at the high end of the consequence
spectrum, Mission (B), then its capability is woefully inadequate and leaves
unmitigated risks.
Notional Cybersecurity Capability Gap

Mission B
Severe
Project power / damage or destroy
Potential
Nation-state / unlimited resources
Unmitigated Risk
Risk Mitigation
Capability
Nation-state / terrorist
Risk
Mission A
limited resources
Project power
Nation-state /
Very Low
Criminal
Hackers
gang
Steal
Threat
Figure 3.
By inference, it is vitally important for national leadership to understand which

threats on the continuum it is most interested in addressing, and receive a
straightforward assessment of national capabilities mapped against those threats. This
is true for individual government organizations as well. Any delta between mission and
capability deserves the focus of decision-makers.
8. Summary
For which discrete categories of threat on the continuum are the defense community,
the security community, and the private sector ultimately responsible? Policy
leadership in any country must make that clear. Only then can operational leadership
ensure that short-term actions and operational objectives measure up against
appropriate long-term strategy.
References
[1] “Defending Against Cyber Attacks”. 2009. North Atlantic Treaty Organization. 9 July 2009
http://www.nato.int/cps/en/SID-D30474D0-A97D6B01/natolive/topics_49193.htm
[2] Bracken, Paul. “Net Assessment: A Practical Guide,” Parameters (Spring 2006), p. 100.
http://www.carlisle.army.mil/usawc/parameters/06spring/bracken.pdf
[3] Box, George E. P.; Norman R. Draper (1987). “Empirical Model-Building and Response
Surfaces”. Wiley, p. 424. ISBN 0471810339. Wikiquote. 9 July 2009
http://en.wikiquote.org/wiki/George_E._P._Box
[4] Gross, Grant. “U.S.Government Focuses on Securing Backdoors in Tech Product”. Security
Central (September 2008). Infoworld. 9 July 2009 http://www.infoworld.com/d/security-
central/us-government-focuses-securing-backdoors-in-tech-products-853
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-77
Cyber Terrorism:
A New Dimension in Battlespace
Major J P I A G CHARVAT
SO2 Course Director
Centre of Excellence Defence Against Terrorism
Abstract. This paper discusses the concept of terrorism, who the terrorists are and
develops an understanding of why they conduct the activities they do.
Understanding the mens rea of the attacker will allow consideration of the type of
attack they may plan and the effect they are likely to try and achieve. It looks at
the main motivations of terrorist groups and discusses their use of the Internet for
various aspects of a terrorist campaign such as propaganda and recruitment. It will
consider the various tactics that have been used and how the Internet has provided
a new opportunity for terrorists to conduct their campaigns and how it has been
adapted by them for their purposes. It examines the potential threat of a cyber
attack by terrorist organizations and how they can use the Internet and Cyber
Space to attack a target with similar results to a conventional physical attack. The
paper will briefly discuss some of the possible defences against this form of
terrorism.
Keywords. Terrorism, Terrorist motivation, Cyber attack, Terrorist use of the

Internet
Introduction
Since the 2001 attacks on the United States there has been a significant effort from
NATO and its partners to address the issue of International Terrorism. Terrorism has
many different types and there is no profile that consistently fits either a terrorist
organization or an individual terrorist. Terrorism is an adaptive threat and will
constantly look for weak points in the state or organization it is attacking. Cyber Space
and the Internet are providing an emerging battleground that many terrorist
organizations are trying to exploit as a means of furthering their campaigns or actual
attacks using an electronic medium. As developed societies become increasingly reliant
on electronic communications, control systems and commerce the potential for a
terrorist to hit the target becomes a more realistic possibility. As with a conventional
attack or command and control medium, the Internet now offers genuine targets that
will become attractive to certain terrorist organizations for certain acts. Just as when
defending against conventional terrorism, cyber systems can be secured and any
potential threat investigated using electronic evidence to catch the instigators.
78 J.P.I.A.G. Charvat / Cyber Terrorism: A New Dimension in Battlespace
1. Defining Terrorism
Most people would say they know what terrorism is, but surprisingly there is no
internationally agreed definition. There are literally hundreds of different definitions in
current use with the use of violence or threat of violence being the only general
common theme [1]. The only other elements to appear in more than 50% of definitions
are “Political” and “fear, terror emphasized” [2]. This is a hamper on international
cooperation as some terrorist organizations are seen as legitimate fighters by some
countries. Terrorism does differ from other crimes in its mens rea; it is done with a
purpose and a specific strategic outcome in mind.
If there has been significant international and intellectual disagreement about a
definition for terrorism itself, the disagreement as to what, if anything, constitutes
Cyber Terrorism is even more diverse. This paper will consider the battle against
terrorism in Cyberspace, Cyber Terrorism as a form of attack and the terrorist use of
the Internet as a tool for physical action. It will also consider areas of both anti and
counter terrorism within a cyber environment.
It is important that we consider who the terrorists are. There are literally hundreds
of groups of varying size and ability, which, to some extent, warrant the label of
terrorists. Terrorism has 4 classic motivations [3]. Firstly there are single-issue
terrorists, those who believe in a particular cause and are prepared to use violence to
protest their message in the hope of ending the issue, which sparks their grievance.
Animal Rights and Anti-Choice over abortion are the two most prevalent of such
issues. Vivisection researchers or family planning workers have been the targets of
sustained campaigns and assassinations in protest over these issues. Although generally
small and with a low lethality rate, these groups could find the cyber world particularly
to their liking as in the cyber environment they can effectively punch above their
weight. Ideological Terrorists are those who use violence to promote their political
ideology, usually from the far left or right. While these groups were most active in the
Cold War, there are still several of such groups still active 1 and other active groups that
have evolved from their beginnings as an Ideological Terrorist group.2
Nationalist terrorists have been the most lethal of all terrorist groups over the last
40 years3 [4] and are still active in several major campaigns worldwide. These are
terrorist groups who seek independence from a state or to cede from one state to
another because of ethnic or geographic grievances. Very few modern nations are made
up of just one ethnic group and in many areas of the world this has led to ethnic
tensions that have spilled over into terrorism. The LTTE in Sri Lanka and the PKK
Kongra/Gel4 terrorist organization in Turkey are the most active of such groups.
Religio-Political terrorist groups tend to be more lethal as they believe they are
acting for God or on a divine order and that those not of their belief are against God
[3]. There are extremist groups spanning all major religions and some minor cults who
have resorted to terrorism. These terrorists have abused their religion and act outside it,
they must not be confused with the religion they misrepresent in their claims. Although
many religions do accept that there are circumstances for justifiable violence or
warfare, none, with the exception of a doomsday cult such as Aum Shinrikyo, would
1
FARC (Revolutionary Armed Forces of Columbia) are probably the most active today.
2
The PKK Kongra/Gel terrorist organisation, active in Turkey, began seeking a socialist revolution in that
country.
3
Religio-Political groups have been more lethal over the last 5 years.
4
The EU lists LTTE and PKK as terrorist organizations.
J.P.I.A.G. Charvat / Cyber Terrorism: A New Dimension in Battlespace 79
apply this to the indiscriminate targeting of civilians or security forces outside the legal
conventions of legitimate warfare.
As with any definition labelling model, there can be hybrid terrorist groups that
either evolve their motivations or have multiple aims. The Provisional IRA are an
example, they were a Nationalist group as they wanted Northern Ireland to cede from
the United Kingdom to the Irish Republic but were also an Ideological group as they
wanted Ireland to become a Socialist state.
The terrorists themselves must be considered, understanding their psychology is
important in understanding how to defeat them. There is no clear profile of a terrorist,
they come from all walks of life and have varying levels of education, employment and
wealth. One common factor is that they are not mentally unstable, terrorist
organizations want activists with the ability to think and be reliable. The level of
intelligence may decide the role of the terrorist, as will any specialist skills such as
chemistry or IT, and the organization will require College level members as well as
those with more basic standards of education. We must accept that most terrorist
groups are made of skilled and intelligent people who are acting out of genuine belief
(self-formed or indoctrinated) and not a group of clueless idiots. This must be
considered in the cyber defence plan against terrorism, they will study, take time, plan
and employ experts of the highest calibre to achieve their aim.
2. Terrorist use of the Internet
Terrorists use the Internet for a variety of reasons that although nothing new per se
make their business far simpler and have a far wider reach than through non-electronic
means. Cyber Space offers many areas of potential exploitation to terrorist
organizations. It is a tool for recruitment, radicalization, propaganda and fund raising as
well as offering quick and simple command and control. Undoubtedly the Internet and
electronic mediums offer many advantages to terrorists, given the ease of use and the
increased reliance of developed societies on the Internet the potential for terrorist
exploitation is expanding daily.
The first area of concern is the terrorist use of the Internet. By this I mean a
medium for many aspects of terrorism other than as an attack tool, which will be
discussed later. There are many features of terrorism that can be conducted through the
Internet, although this is primarily an information and communication medium it does
allow a greatly extended reach and far quicker communication. Terrorism is a
politically motivated act and therefore requires publicity and a forum of
communication. The key areas of exploitation in modern terrorist use are propaganda,
recruitment, radicalization, communication and research. The Internet allows small
groups or individual terrorists the opportunity to reach literally millions of people very
easily. “Perhaps one of the most promising features of the Internet is that it gives voice
to many who have been unable to buy or generate media attention” [5].
Propaganda is essential to terrorism, these are rational people who carry out their
campaigns for a political or social end state, they must have a medium to explain their
message and ‘justify’ their actions. Before the Internet this was relatively difficult to
achieve for a mass audience. Television and print media would run stories on terrorists
but these would be subject to editorial control and sometimes to legal restrictions.
Books, magazines and pamphlets would only hit a small audience and would only
really be read by those with an interest in the terrorist’s cause. The World Wide Web is
unregulated and accessible to almost everyone. Internet Cafes are increasingly popular
and allow access to the public without the need for the ownership of a computer or
subscription to an Internet Service Provider. Simply by adding the correct key words or
links via a seemingly mainstream site, the terrorist organization can easily display its
message without regulation. It can allow the key grievances, which motivate the
terrorists, to be publicly aired, and as they control the content, these can be backed up
with ‘proof’ manufactured and edited by the organization. These sites look official and
will use multimedia to attract attention and create an air of legitimacy for the site
visitor. They also have the advantage that their Information Operations are not bound
by truth or conventions, which allows, with appropriate editing, a seemingly
convincing piece of propaganda without any real basis. By May 2005, using only the
US State Department’s list of terrorist organizations, there were over 4500 terrorist
supporting websites [5], rising to 5500 in 2007 [6].
An example of how easily websites can be used for misinformation is the German
Bund Deutscher Juristen[7]. This website for the German Lawyers’ Association ran an
article about their Chairman, Dr Claus Grötz, quoting that he said the possible use of
testimony gained after light torture could be used in German courts. There was public
outcry and headline news calling for his resignation. The site had only been set up 2
days before the ‘story’ broke and neither the Association nor Dr Grötz actually existed.
Although this is not a terrorist example, it highlights the possibilities that a terrorist
organization could exploit. Lazy journalism meant that the story was not corroborated
and given to the German public as authentic by the mainstream media. To have this
about a State’s action relating to a terrorist situation could be used by the terrorist
organization to gain public sympathy and international condemnation of the victim
state.
We have to break
taboos these days. The
gaining of testimonies with
the help of light torture
and the utilization of such
testimonies before court
have to be made possible in
the near future.
Dr. Claus Grötz, Chair-
man of the German Law-
yers’ Association and cri-
minal judge at the German
Federal Court of Justice
Figure 1. Screenshot of BDJ website and translation
Recruitment and Radicalization are an essential element for a terrorist

organization. The Internet provides a greatly enhanced forum for this. The ability for
terrorists to find and groom young people is demonstrated in Forum and Chat Room
websites. This provides a largely unregulated medium for terrorists to meet and groom
potential recruits. Often they will monitor those Forums and Chat Rooms that may
have a relevance to their motivation, grievance or cause. This could be an animal rights
Chat Room where extremists Single Issue terrorists may use the opportunity to engage
anyone who shows thoughts or emotions along the same lines. This form of contact can
be well orchestrated and involve several people, effectively keeping the potential
recruit in an air lock away from the terrorist proper until they are deemed ready.
Initially a pro-terrorist ‘chatter’ will engage the potential recruit in fairly mainstream
conversation and ask a few probing but seemingly innocent questions over a period of
time. They will use this time to pass on pro-terrorist messages and try to affirm the
potential recruit’s feelings to that particular cause. The will also post messages against
the terrorist’s targets in an attempt to convince the potential recruit that the terrorist
message is accurate.
Once regular contact is made, the initial contact will assess the potential recruit
and pass them on to a groomer. Those selected for such grooming have already
displayed some form of agreement with the terrorist cause and also the personality
traits that suggest they may be willing to take an active part in any struggle. Still at this
point the potential recruit is unlikely to know she or he is in contact with anyone other
than a fellow chatter of a like mind and engaging in serious conversation. The groomer
is likely to be very knowledgeable about the cause and will start to feed strong
propaganda about the terrorist’s motivation. This is again a filtering process to find out
those who would continue towards direct action from those who have strong views but
would never cross from political protest. Those who are regarded as strong believers in
the terrorist cause, and displayed the correct personality traits to suggest they would
join are then passed on to their first proper contact with the terrorist organization itself.
This process can take a long time, as the groomer has to be certain they have the right
people and the potential recruit is not going to be made aware that he or she is in
contact with a terrorist until they are ready.
The groomer will pass the potential recruit onto a recruiter who, at that stage will,
for the first time, make indications that they are from a terrorist organization. From this
point the skill set of the potential recruit will be examined and their commitment finally
checked before they are in a position to ever actually meet or know the identity of
anyone they have been engaged with.
An example of this, in its early stage was monitored on a mainstream Muslim
Youth website in the United Kingdom.5 Hussain was a 15-year-old schoolboy who
posted a school project for comment from other members of the forum site. In his post
he expressed mixed feelings and uncertainty about how the West saw Islam and the
true nature of the Jihad. Hussain stated that he believed Jihad was a personal struggle
and it was against Islam to kill. He also expressed that he felt as though the West
regarded him as a terrorist because he was Muslim and that there was significant anti-
Muslim sentiment in Western Society. The first two replies agreed with Hussain that in
Islam it is forbidden to kill innocents. OBL4Caliph entered the debate and began
saying that he was an authority on Islam and that Jihad was a duty for all Muslims and
that it was a requirement to kill those who opposed the religion. During the ensuing
posts it was clear that most forum members said OBL4Caliph was wrong. However his
language and argument were more structured to a youth’s mind and he began to try and
convince Hussain. While clearly Hussain had used the Internet for a sensible and
5
This case was monitored by the author, the website name is withheld as it has no association with terrorism
and quickly banned OBL4caliph from the site. There is no evidence OBL4Caliph directly represented or was
a member of any terrorist organization.
reasonable purpose, canvassing views of like-minded people about his thoughts as a

confused teenager, he had inadvertently shown a little potential in his thought, which
led to a potential terrorist grooming.
Terrorism has evolved and in the beginning of the 21st Century we are dealing with
a new type of terrorist organization as well as the classical groups. Traditionally
terrorist organizations have been just that, an organization with a leadership and strict
control. Attacks and campaigns would be planned and authorized by the leadership as
part of a coordinated approach to their policy. The emergence of more ‘networked’
organizations with a horizontal leadership has made the Internet a breakthrough in
Command and Communication. Obviously e-mail is an instant form of communication
and can easily be encoded. Seemingly innocent messages can be sent that only the
recipient would understand, although a coded message is as old as terrorism itself,
email allows much greater speed in delivery and an almost guaranteed receipt. An
example was a mail sent by the 9/11 hijacker Mohammed Atta:
“The semester begins in only three more weeks. We’ve obtained 19
confirmations for students in the faculty of law, the faculty of urban
planning, the faculty of fine arts and the faculty of engineering.
Best wishes from the Professor to all of you!
Mohammad” [7]
Clearly this message seems innocent, but knowing the events of 9/11 and who
wrote it, we know this set the attack dates (it was written 3 weeks before the attacks),
the 19 confirmations were the 19 terrorists who were carrying out the atrocities, the
‘faculties’ were codes for the targets and this confirmed which ones, and of course the
‘professor’ is Osama Bin Laden.
The Internet also offers a unique opportunity for people to meet likeminded others
despite geographical separation. Via forums, chat or other emerging methods such as
Twitter or Facebook, unconnected and uncommon ‘radicals’ could easily contact other
uncommon radicals and find a common virtual space online.
Messages can be hidden in pictures or a made to look like Spam mail so not to
attract attention. Terrorists have also been known to set up an email account and
change the password daily, the cells and terrorists will know the username and daily
password. Messages can be written, saved as a draft and then accessed by the whole
network without being sent. This greatly reduces the possibility of interception or an
evidence trail before or after an attack.
Simple symbolism is also a known method of terrorist communication. Many
terrorist organizations have websites or will publish video clips on video sharing sites.
These can contain a hidden code; simple graphics such as a terrorist holding his or her
AK47 in the left hand will be displayed. Having the same graphic with the rifle in the
right hand can be a signal for a terrorist cell and be almost undetectable to the
intelligence services monitoring it.
Training and Research & Development are essential elements for a terrorist attack.
The Internet provides easy and immediate information sharing and research capabilities
to a Terrorist organization. There have been numerous examples of terrorists posting
training manuals on the Internet, which explain how to conduct attacks and make
explosives using readily available high-street ingredients. This has a greater appeal to
network organizations with horizontal hierarchies, such as al-Qaeda. These
organizations would be content for cells, even with no formal contact with the
organization proper, to conduct attacks in their name. Their philosophy is espoused on-
line and those who are radicalized to support it could become competent terrorist
without physically going to a training camp. Publications such as the Terrorist
Cookbook provide the know-how that a budding terrorist would need.
Fund raising for or by a terrorist organization is another area where the Internet
provides a quick and simple medium for the organizations. In the modern world of
electronic banking this can be achieved directly or indirectly, both via legitimate
transitions and illegal means. The Internet offers many opportunities for front business
and pseudo-charities to raise monies. It also allows easy transfer, internationally, to
make the tracking and freezing of suspect terror funds very difficult. In some examples,
charities have been set up for disaster relief, such as the devastating earthquake in
Pakistan in 2005. While these charities have done some relief projects, some of the
funds were siphoned off to terrorist organizations. If $100,000 is set aside to build a
school, the actual relief could be $800,000 with 20% going to terrorism. These
‘charities’ can be easily set up by terrorist supporters and be indistinguishable from
genuine relief work. The Internet allows a wide target audience for donation requests
and easy transfer to the ‘relief’ fund. At a time of such disaster considerable amounts of
money can be stolen in this way.
3. Terrorist Cyber Attack
There are many who argue that there is no such thing as Cyber Terrorism proper.
Terrorists can use the Internet as discussed above, but Terrorism needs to be a tangible
physical attack. I would disagree, there are many features of modern life that are reliant
on Cyber Space and present a new opportunity for terrorist direct action, and these
opportunities are increasing as we become reliant on computers. There are the
possibilities of attacking electronic means such as web-defacement, malware, data
mining, training and Denial of Service. As SCADA becomes more widely used and
controls important key infrastructure, an attack on these could be as physical as a
bomb.
Web-defacement is an easy way to annoy a target or gain propaganda. Unlike a
spurious site, such as the BDJ, these attacks alter the data and information on an
official site. The ISP and web domain name would prove to be official if they were
checked.
While simple altering of an official website gains little more than small scale
propaganda, there is a potential to cause real panic. The often given scenario is a
terrorist defacing the Homeland Security website and advising people to leave a major
city due to a chemical leak. If picked up by the press or enough people, this false
message could gain legitimacy and cause panic, with untold casualties in the ensuing
rush to leave and the obvious financial implications that would cause. This sort of
attack is relatively unlikely to succeed as other forms of warning would not back it up,
but if a TV network accidentally picked it up it could be damaging. The increase in
popularity of ‘New Media’ does allow a greater potential for this type of scenario as it
as unregulated and can lack the responsibility of confirming a source that traditional
media would have. Social contact media such as Twitter could inadvertently spread
false information and be taken as genuine. Although most users are responsible and
corrective information will quickly appear a botnet giving mass ‘tweets’ may give
legitimacy through weight of numbers [8]. This is a very simple action for a skilled
hacker Cyber Terrorist with very little risk or cost.
Malware is an obvious weapon that a cyber expert terrorist could unleash. Viruses
and works can bring down systems and networks and cause great disruption to the
target. These could render an important operating system temporarily useless or make it
malfunction. The potential loss of data through such a virus or work could have a huge
implication if targeted correctly.
Data mining is an appealing prospect for a terrorist organization and an area where
INFOSEC becomes a priority for governments and security services. Increasingly
personal and financial details are held on record in electronic files. Often, for ease of
use and legitimate information sharing, these are held on networked systems. Terrorist
organizations could attempt to hack in to these systems to gather information about
potential targets, financial details or indeed information altering to damage the victim
organization. This could be used to identify key individuals to target for assassination
or kidnap. It could also find details, which it used to discredit or blackmail key
personnel to help with a terrorist activity. Identity theft could allow a terrorist access to
bank, identity documents or access control passes which could be severely damaging
and greatly assist in an attack. Given the ease and size of modern data transportation
mediums, such as flash USB sticks, the loss of this information must be guarded
against in all electronic forms. In the United Kingdom the membership list for the right
wing British Nationalist Party was made available on the Internet, much to the
embarrassment of several high profile members whose membership of this party was
proscribed by their employers6. There are many such types of information, which could
be of use to a terrorist organization. They can also learn about the schedules and
locations of targets. According to an al-Qaeda training manual captured in Afghanistan
“Using public sources openly and without resorting to illegal means, it is possible to
gather at least 80% of all information required about the enemy”[6].
Seemingly innocent programmes and tools on the Internet can provide valuable
information to a terrorist planning a physical attack. These are in everyday use by
millions of Internet users for wholly legitimate purposes. Programmes are easily
available that offer satellite imagery of the world. These offer exact details of often-
sensitive areas and can allow accurate target selection and area knowledge without the
risk of reconnaissance. Indeed the al-Qaeda computer expert Muhammed Naeem Noor
Kahn was captured in Pakistan in July 2004 with a computer filled with photographs
and floor diagrams of buildings in the US [8].
If the Internet is brought down in a country or region it will do immense damage to
the population, infrastructure and financial sectors of that society. The Internet relies on
bandwidth and if targeted with a Distributed Denial of Service, DDoS, this can block
the selected servers and effectively jam the Internet. This is relatively simple to do. If
enough emails are sent in a short enough time, or enough hits are made simultaneously
on a selected website, the bandwidth will fail to cope with the amount of data it is
being requested to move and simply clog up, rather like a main street in rush-hour. One
method of doing this is through a Botnet. A Botnet is a network of computers that have
been taken over as slaves by one master computer. From the master computer a
terrorist could direct all the Bots in the Botnet to email or log onto a website
simultaneously. A Botnet can be millions of computers all over the world with the
6
Clearly this is a non-terrorist related example and no inference relating the BNP to terrorism is drawn in any
way.
slaves showing no apparent symptoms. Botnets are relatively easy to set up for an
experienced hacker and are available for purchase. Although not a terrorist attack,
Estonia was the victim of a sustained DDoS attack in 2007. As Estonia prefers to
operate as paperless as possible, effectively closing the Internet had a huge effect in
that country.[7] This was combined with some civil disturbance relating to the moving
of a Soviet war memorial in Tallinn city centre. This could be used by a terrorist
organization to bring down a major banking system, Bishop’s Gate and the Baltic
Exchange bombs in the early 1990’s in London highlight the attraction of a purely
economic target to a terrorist organization. What the Provisional IRA achieved with a
bomb could now be done to a much larger target area with only a laptop.
Supervisory Control And Data Acquisition (SCADA) systems control many major
infrastructure systems and are increasingly being relied upon. These provide one of the
greatest vulnerabilities to a purely Cyber Terrorism attack with potentially massive
physical effect. SCADA systems are used to run power plants, control dams and even
city traffic flow by controlling the traffic light system. If these can be accessed by a
terrorist organization they can effectively take control of that facility. One of the main
vulnerabilities for this is through a mole employee or a disgruntled worker. If a terrorist
organization placed members as regular employees of a facility it is possible that they
could, given the patient nature of sleeper cells, gain a position of trust in the facility
and gain access to the SCADA computer system. From there they could initiate an
attack to break a key facility or cause other forms of damage. An example of the
potential this has happened in Queensland in Australia. Although not a terrorist attack a
hacker got into the Sewerage SCADA in Maroochy Shire Council on Australia’s
Sunshine Coast. Vitek Boden put in glitches and deliberately released millions of liters
of raw sewerage into the water system and sparked an investigation. He was a former
employee who had access to the required passwords and knew the system. After his
dismissal the council had failed to change the passwords and effectively allowed Boden
access to the system [8]. His motives were personal but this shows what a terrorist
could do.
4. Conclusions
The face of modern terrorism is ever changing and it seeks new methods of carrying
out attacks, propaganda campaigns and recruitment. Cyber Space in certainly a new
area of a battlefield and one that terrorist organizations are striving to exploit. There are
many advantages to the terrorist to use the Internet for a myriad of essential threads to
maintain a terrorist campaign. This threat must not be overlooked, as many societies
move more areas of life and infrastructure to computer control and networked systems,
the reliance on then is ever increasing. Unfortunately this reliance creates a fairly soft
target. The terrorist no longer needs to physically be in the same country, let alone the
site of an attack if it is conducted through cyber space. Information is far more
accessible and available instantly, something a terrorist could exploit. Taking a
photograph of a military instillation will raise suspicion and risk alerting the authorities
to the terrorist or that an attack is being planned. Looking at the same site on a
computer would leave the terrorist completely anonymous and undetected.
Terrorists such as Younis Tsouli, the infamous Terrorist 007, have become vital to
terrorist organizations. Tsouli was not a fighter that offered much ability to conduct a
physical attack, but he was committed to the al-Qaeda cause. He set up countless
websites and video sharing forums to promote a pro-al-Qaeda message and demonize
the US and UK. He operated from a London flat and was financed by another terrorist,
interestingly who he never actually met. Tsouli was arrested with millions of files and
videos with terrorist propaganda and training aids, which he was the central hub in
distributing and putting on line. It was assessed that his arrest as a major blow for al-
Qaeda, as much as any active field commander.
It is not all bad news though, by using electronic means, the terrorists can leave a
signature and be monitored or arrested based on electronic evidence. It can also be used
to monitor terrorist ‘noise’ [9] as an intelligence-gathering tool. Like any form of
attack, a cyber attack is likely to leave some form of signature or evidence that if
properly monitored or collected can be used as a counter terrorist tool. There is also
some potential to use cyber means to attack the terrorists back, however this has some
ethical dilemmas and is not within the scope of this paper.
Defence Against Cyber Terrorism will differ little form solid cyber defence and
security. It is however important to understand the terrorist mindset and how they are
likely to use the medium for their purpose. It must be conducted as part of the wider
force protection and be conscious of the massive potential for compromise. It is highly
unlikely anyone would take a large paper file, with highly sensitive information useful
to a terrorist, outside a secure office, but this seems to be ignored when that
information and a thousand times more is held on a 5cm USB stick. It is not only the
system that requires improved security if the defence against Cyber Terrorism is to be
successful. The often used comment, TPIBKAC, The Problem Is Between Keyboard
and Chair and that there is no patch for stupidity hold true.
Having attended several workshops on Cyber Defence, one area of concern is that
many ‘experts’ believe Cyber Terrorism simply won’t happen. It is foolish to write it
off, at best it may currently be more ‘potential than problem’, but it is short sighted to
exclude it from risk assessment. Firstly, it may have already, we don’t know who may
be in place ready to attack a SCADA system, secondly we are increasing our reliance
on these systems and their attractiveness as a target grows. Without doubt, the modern
terrorist needs the Internet as much as the AK47 and it is a factor we would ignore at
our peril.
All views expressed in this article are the authors and do not necessarily reflect or
represent the views of COE DAT, CCD COE, NATO or the UK MOD or Government.
References
[1] Record, Jeffery: Bounding the Global War on Terrorism, Strategic Studies Institute, US Army War
College, Leavenworth, 2003
[2] Schmid, Alex and Jongmans, Albert et al: Political Terrorism: A new guide to Action, Authors,
Concepts, Data Bases, Theories and Literature, Transaction Books, New Brunswick, 1988
[3] CSTPV St Andrew’s University Certificate in Terrorism Studies
[4] COE DAT Information Collation Managemant Cell database
[5] Weimann, Gabriel: Terror on the Internet, USIP, Washington DC, 2006
[6] Weimann, Gabriel: WWW.AL-QAEDA: The reliance of Al-Qaeda on the Internet7
[7] COE DAT Cyber Terrorism Couse IV Mar 09
[8] COE DAT Strategic Communications Workshop May 09
[9] Huizing, Harry: Cyber Terrorism Briefing Note, COE DAT, Ankara, 2008
[10] Krone, Troy: Gaps in cyberspace can leave us vulnerable, Platypus Magazine (edition 90, Mar 2006)
7
Thanks to Prof Weimann for his kind permission to use this article.
[11] COE DAT Cyber Terrorism Workshop Oct 07

[12] Bunker, Robert J: Networks, Terrorism and Global Insurgency, Routledge, Abingdon, 2005
[13] Hennessy, Joh L and others: Information Technology for Counterterrorism, National Academies Press,
Washington DC, 2003
[14] Hoffman, Bruce: Inside Terrorism, Columbia University Press, New York, 2006
[15] Huntington, Samuel: The Clash of Civilizations, Free Press, London, 2002
[16] Laqueur, Walter: The New Terrorism: Fanaticism and the Arms of Mass Destruction, Oxford
University Press, New York, 1999
[17] Sageman, Marc: Understanding Terror Networks, Penn, Philadelphia, 2004
[18] Stern, Jessica: The Ultimate Terrorist, Harvard University Press, Cambridge MA, 1999
[19] Tuman, Joseph S: Communicating Terror, Sage, Thousand Oaks, 2003
[20] Whittaker, David (ed): The Terrorism Reader 3rd Ed, Routledge, London, 2007
[21] Wilkinson, Paul: Terrorism Versus Democracy, Routledge, London, 2006
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-88
Borders in Cyberspace: Can Sovereignty

Adapt to the Challenges of Cyber Security?
Forrest HAREa,1
a
School of Public Policy, George Mason University
Abstract. The new US administration has begun efforts to securitize the

substantial problems the United States is currently facing in cyberspace. Recently,
President Obama ordered his National Security Council to conduct a rapid review
of existing measures being undertaken by the federal government, and provide
recommendations for additional ones. Many stakeholders in the US government
and private industry are watching these actions closely as there seems to be broad
acceptance that the issues call for more extensive security measures. However,
many issues will complicate effective securitization of threats in cyberspace. For
example, not all stakeholders agree on the priorities or where the focus of security
measures should be yet cyber security is a “trans-sovereign” issue affecting both
developed and developing countries in an interdependent manner.
Because actors in cyberspace enjoy relative anonymity and can threaten inter-
connected targets around the globe, there is considerable debate as to whether the
concept of borders is relevant to the challenges of cyber security. Regardless the
focus of the debate, the concept of borders is important because they define the
territory in which national governments can employ sovereign measures. To
analyze borders in the context of cyber security, this paper asks the question, “Is
there an important role for the concept of borders, if not physical lines, in
improving national security in cyberspace?” To explore the question, the paper
takes two approaches. The first is a comparison of the cyber security issues to
international drug trafficking in an effort to explore how sovereign measures used
to combat drug trafficking may be applicable to improving cyber security. The
second approach is an examination of the issue from the perspective of the Heal
and Kunreuther Inter-Dependent Security Model with an attempt to inform the
cyber security decision process of national governments as they consider options
to invest in a higher level of security.
The paper will argue that, whether the problem is addressed from the standpoint of
criminal behavior like drug trafficking, or cyber attacks in an interdependent,
global domain, borders can be a potentially useful construct to address cyber
security issues and inform national policy decisions, regardless of the physical
location of relevant nodes. However, sovereign powers must be careful not to use
the concepts of borders to curtail the progress our nations have made to connect
and better the world via this evolving and expanding environment.
Keywords. Cyber Security, borders, agent-based modeling, interdependent

security model
1
Corresponding Author: Forrest Hare; E-mail: fhare@gmu.edu
F. Hare / Borders in Cyberspace: Can Sovereignty Adapt to the Challenges of Cyber Security? 89
Introduction
In the United States, there have been multiple initiatives to raise awareness and
securitize the nation’s vulnerabilities in the medium of cyberspace2. Many stakeholders
in the US government and private industry are watching these actions closely as there
seems to be broad acceptance that the issues call for more extensive security measures.
Two initiatives are noteworthy- the Congressional Commission on Cybersecurity for
the 44th Presidency, and the Obama administration’s 60-day cyber review. The
commission’s report [2] identified cyber security as “one of the major national security
problems facing the United States (pg 1).” In addition, President Obama ordered his
National Security Council to conduct a 60-day review of existing measures “to ensure
that U.S. Government cyber security initiatives are appropriately integrated, resourced,
and coordinated with Congress and the private sector [3].”
In many cases, for example in response to the recent violence on the U.S./Mexico
border, an important component of any security response is often a call to “defend the
borders”. However, such measures are problematic when attempting to respond to
threats in cyberspace. One of the biggest issues with applying border control measures
in cyberspace is the amount of inter-connectivity with other nations in the medium.
Because cyber security measures must be internationally coordinated, the question
often arises as to whether the concept of borders is relevant in the domain [see 4,5].
This article explores the question, “can borders, as components of sovereignty, be
relevant to addressing cyber security?” To explore this question, I will use two
different analytical constructs. First, I will compare the problems with securing a nation
in cyberspace to the problems of combating drug trafficking. Researchers have raised
similar concerns about stemming the tide of illegal drugs crossing national borders. If
these two problems appear similar in their challenges, then perhaps we can draw
lessons for cyber security from border-related measures to combat drug trafficking.
Second, I will apply the Interdependent Security model, on which I will elaborate in a
later section, to the problem of national cyber security. If this model can be considered
a valid construct, perhaps it can also point to a role for borders as the nation-state actors
in the model choose cyber security investment strategies. At a minimum, it would
highlight in which countries sufficient measures are being taken and where they aren’t,
thereby highlighting the boundaries between them. Based on these dissimilar analyses,
I will argue that, regardless the challenges of applying security measures “at the
borders,” as concepts of sovereignty, national borders remain relevant components of
state-level responses to security threats in cyberspace. This analysis will not be an
attempt to find answers to empirical questions, but rather provide new frameworks
beyond the purely technical/legal aspects to address cyber security and borders.
The next section will provide some background and frame the discussion more
precisely. There are several challenges to even effectively securitize threats in
cyberspace. For example, the nature of “cyber security” as a national security issue is
ambiguous and there is a heightened potential for a security dilemma in the domain.
The goal of this section will be to frame the problem in such a way that it effectively
bounds the ensuing discussion and informs the questions of borders as they relate to the
2
For the purpose of this article, “securitization” is understood as the process outlined by Buzan et al. [1].
Namely, an issue is presented as posing an existential threat to a designated referent object (in this case, a
nation-state) requiring emergency measures and justifying actions outside the normal political bounds (pgs
23-24).
90 F. Hare / Borders in Cyberspace: Can Sovereignty Adapt to the Challenges of Cyber Security?
domain. After addressing these issues, I will present the two analytical frameworks and
highlight their relevant findings. The article will conclude with a synthesis of the issues
and a discussion of policy implications. Regardless the findings from this work,
sovereign powers must be careful not to use the concepts of borders to curtail the
progress we have made to connect and better the world via this evolving and expanding
environment.
1. Framing the Issue
Effective securitization of threats in cyberspace can be complicated by many issues.

First, there is little agreement as to what the security issue in cyberspace actually is.
This is a common problem with issues of security that must compete to be on the
public agenda. Arnold Wolfers [6] called national security an “ambiguous concept,”
and because of the unknown nature of actors and their motives in cyberspace, the
ambiguity is only heightened in this domain. Different actors will securitize the
problems according to their perceptions or agendas. For example, while one nation may
assert that an existential threat is posed by a denial of service attack against their fragile
banking infrastructures, another may highlight fundamentally different issues. Some
policy advocates would include threats from websites critical of government regimes to
be a component of cyberspace security. A. Strelstov [7], a member of a Russian
delegation to the UN, identified that, “undermining a state’s economic and social
systems and psychological manipulation of a population for the purpose of
destabilizing society,” is also a component of what the Russians call international
information security (pg 8).
An additional concern is the heightened potential for a security dilemma in
cyberspace. As characterized by Herz [8], a security dilemma may arise as one nation’s
efforts to arm themselves in defense may provoke another nation to do likewise,
thereby creating a greater threat. Buzan [9] goes further to identify that some
ambiguous measures may actually be attempts to gain more power vis ¥ vis potential
adversaries. This challenge of what Buzan terms the “power-security dilemma” is most
difficult to counter in cyberspace. When fielding tanks and anti-aircraft missiles, their
presences can be declared as defensive measures and made visible to the public.
However, it is much more difficult to make public or confirm the defensive nature of
measures a country may employ to improve security in cyberspace. Assertions that the
actions are also offensive will be difficult to counter because any offensive potential
would be difficult to disprove and offensive use would be difficult to identify.
Complicating the issue further, attacks within in the domain can easily be masked and
attributed to a nation-state, when, in fact, they may be the actions of non-state actors (or
vice versa).
Overcoming both these factors requires a common understanding of the issues.
Any effort to securitize a situation requires a threat agent, a victim, and an
understanding of how the threat agent causes an existential threat to the victim. In
cyberspace, the threat agents can be criminals, hackers, terrorists, and nation-states 3.
The greatest challenge is determining who is conducting the attack before extensive
forensics have been accomplished. The targets of these actors are also diverse. They
3
For a good survey of the myriad of malicious actors in cyberspace, see Denning [10], Gorman [11], and
Kramer et al. [12].
may be in the business of stealing personal identities to commit fraud, conducting

industrial espionage, engaging in cyber extortion of critical infrastructure owners, or
preparing for and conducting a deliberate conflict accompanied by actions in
cyberspace. Any analysis and body of policy recommendations that attempt to
incorporate every possible combination of these malicious actors and their attack
methods would be hard pressed to escape the trap of ambiguity. Therefore, to narrow
the scope to a level appropriate for this analysis, national-level cyber security will
entail the following:
Attacks and infiltrations by either state or organized non-state actors against
government and critical infrastructure systems (privately and publicly owned)
to gain knowledge of a national security value and/or attempt to
degrade/disrupt such systems.
National security is about existential threats to the state. Obtaining knowledge of a
national security value can create an existential threat by allowing potential adversaries
to gain the knowledge to develop effective counter-measures to a nation’s advanced
military and other defenses. In addition, cyber attacks that degrade the ability to
command and control national security assets and attacks that disrupt critical
infrastructure have direct implications to national security. This infrastructure may be
civilian, military, or both. In the United States, for example, the Department of Defense
relies heavily on the nation’s public and private cyber infrastructure backbone for
communications purposes [13].4
Some security measures are currently in place to protect against the threats
articulated above. Such measures are employed by both government agencies and the
private sector owners of much of a nation’s critical infrastructure [see 14]. An obvious
measure to defend against the theft of sensitive information would be to place all
critical information and correspondence on closed systems that are not connected to the
publicly accessible Internet. In the United States, for example, this would entail
containing the information within the national security system architecture managed by
the National Security Agency and Defense Information Systems Agency5. Certainly,
governments secure much of their critical information in this manner. However, it is
also the case that, as we become more reliant on the Internet for collaboration on all
activities, especially between the public and private sector, it is becoming increasingly
difficult to keep critical information controlled in this manner. A recent incident
regarding a potential loss of design information for the F-35 Joint Strike Fighter
highlights this problem. The information was stolen from private, proprietary industry
networks (meaning no government access or frequent auditing), and it apparently
contained several terabytes of design data on the future air defense capability for
several nations [15]. Remaining disconnected from the greater cyberspace could be a
measure employed by critical infrastructure owners and operators also. The control
networks could be closed, proprietary systems with no remote access. In fact, older
generation control systems employed tailored protocols and were only managed
through proprietary, closed systems because there was no Internet available at the time.
4
Note that the focus for this article does not include industrial espionage unrelated to national security,
hacking for pleasure, identity theft, and the use of the Internet for training, messaging, and internal
transactions of bad actors. Though these can all be considered criminal acts in their own right, they are
outside the scope of this discussion.
5
For an overview of the U.S. National Security System, refer to the CNSS website at www.cnss.gov
However, the trend has been to install remotely maintained systems employing
common OS architectures to leverage the connectivity benefits of the Internet [16].
Therefore, these critical infrastructure systems have assumed a risk common to all
those dependent on the effective functioning of the Internet.
The United States, as a sovereign country, certainly has the inherent right to
control all of its borders in any domain [17]. With the above considerations, it is clear
the public sector cannot manage all necessary security actions alone. Private companies
are an important part of the dynamic that is absent in other areas of national security
where the actions of the military, or law enforcement, dominate the response options.
We have no early warning radar system or Coast Guard to patrol the borders in
cyberspace. Unlike in other domains, information of an attack will come first from
those being attacked. Therefore it is highly unlikely that a government organization,
unless it is actually the target of a cyber attack, will have greater situational awareness.
An effort must be made to incentivize the private sector to invest in cyber security as
well. In many cases, national security depends on it. But if none of the measures being
employed have a border patrol component, does that necessarily mean that borders are
not significant in cyberspace? The next two sections will introduce two different
frameworks to address this question. In the first of the two analytic frameworks, I will
compare the problems of securing a nation against cyber threats to the challenges of
securing a nation against international drug trafficking.
2. Comparison with Drug Trafficking
In a way, the world has become a victim of its own developmental successes. Over the
last two decades, we have seen an incredible amount of openness in commerce and the
exchange of ideas. However, with openness comes much vulnerability. Several authors
have highlighted the fact that increased globalization and economic interdependence
have been accompanied by greater economic disparities. Globalization has also created
an environment where it is much easier for clandestine transnational actors (CTAs) to
operate [see, for example 18]6. As such, it has been increasingly difficult to secure
nations against the growing, non-traditional threats from these CTAs. However, it is
possible that efforts to do so can inform the challenges of cyber security. But how
similar are the issues?
From my perspective, there are at least six factors in combating drug trafficking
that compare to the challenges of cyber security. First, CTAs engaging in the drug trade
are based in countries with few legitimate economic opportunities [18]. Legal activities,
such as growing subsistence crops have little chance of competing with the lure of
income from growing poppy seeds in countries like Afghanistan and Thailand. A
similar trend has been developing in cyberspace. Several developing countries have
become sanctuaries for cybercriminals, or transit points for malicious actors located in
other regions. A recent arrest in Romania highlights the growing hacking community in
Eastern Europe. A hacker nicknamed “Wolfenstein” is suspected of breaking into US
Department of Defense computer systems and planting malware [20]. Brazil has also
been identified as a growing contender for the cyber crime capitol of the world. In 2004,
the Brazilian federal police claimed that it was home to 80% of the world’s hackers
[21]. In addition to these countries, researchers consider China to be a growing threat
6
The term “clandestine transnational actors” is adopted from Andreas [19].
[22]. But due to problems of attribution, it is difficult to tell if the actions are state-
sponsored, or private actions [23]. Regardless the location, or identity of a specific
attacker, a nation-state may be involved as the sponsor. Many countries would be
interested in information about rivals thereby acting as customers to those providing
national security-related data obtained through cyber espionage.
Second, but closely related to the previous point, illegal drug control regimes rely
almost exclusively on the coercive actions of national governments, but the trade is
conducted in areas where actions at the state level are often ineffective [24,18].
Because each country is sovereign and reserves the right to draft and enforce it’s own
laws, international drug control actions must contend with widely differing legal
regimes. Compounding the problem, developing countries have much less effective law
enforcement. The same barriers confront cybercrime responders. For example, as
recently as 2004, hacking was not considered a criminal offense in Brazil [21]. Only
recently have such vehicles as the European Cyber Crime convention and other bi-
lateral agreements led to improvements in synchronizing legal regimes to combat cyber
offenses.
Third, the Internet itself is an excellent source of knowledge on how to engage in
the drug trade [24]. One can easily find instructions for how to make such drugs as
LSD or methamphetamine (see, for example, www.homemadedrugs.net) by browsing
websites or conversing with others in forums. Just as easily, one can find the tools
required to break into computer systems, as well as instructions for their use in news
groups (see, for example, www.sectools.org). During the conflict between Russia and
Georgia, for example, there was substantial evidence that the attacks on Georgian
governmental websites were directed via web forums [25]. In neither the case of drug
trafficking nor hacking is formal training required or even available. The only actors
who may have received formal training would have done so as former security officials.
Fourth, customs agents have to sift through ever increasing amounts of legitimate
goods and people to find illegal drugs. According to Stephen Flynn [26], a border
security expert and former US Coast Guard officer, customs agents must patrol a
continuous stream of peoples and goods at more than 3,700 terminals at over 300
points of entry. As he states, it “[i]ntercepting the ripples of danger in this tidal wave of
commerce is about as likely as winning the lottery (pg 57).” Similar challenges exist in
cyberspace. With well over 3 Tbps traversing international routes between the US and
the rest of the world, it is virtually impossible to differentiate legitimate Internet traffic
from traffic with a malicious purpose 7 . Information that has been stolen from
somewhere, or that contains commands that will “flip a switch” in such a way as to
cause severe damage to a critical infrastructure system, is extremely difficult to identify.
Intercepting it requires previous knowledge that the information should not be traveling
across the Internet in the first place. In other words, you can train a dog to identify
marijuana, but it is unlikely it can be trained to identify the difference between Bayer
aspirin and a generic or that the prescription has expired and belongs to someone else.
Fifth, efforts to combat retail drug transactions are constrained by civil liberty
concerns. Victims, who could be considered accomplices in an illicit transaction, can
hide behind privacy rights [24]. Oftentimes the victims of cyber espionage may choose
to cover the event as well, but for slightly different reasons. Since cyber crimes can be
7
3 Tbps is an educated guess based on analyzing the Telegeography data source used for the second analysis
of this paper. An exact figure would be extremely difficult to obtain and would only be valid for a few
seconds.
hidden from the public by both the victim and the perpetrator, a company that has been
infiltrated may chose not to report an event for fear of assuming liability for the actions
[27]. They may also be concerned about a reduction in customers’ trust in their ability
to safeguard sensitive information. In fact, it may not be until the recipient nation of
stolen data has built an exact replica of a system, for which they have obtained the
design secrets, that there is any indication that a theft has occurred.
Lastly, there is little agreement on what exactly constitutes the “evil to be
eradicated” when assessing and implementing counter-drug trafficking measures [24].
For example, the current debate surrounding the drug cartel violence in Mexico centers
on the role of the United States in creating the problem [28]. What is worse, trafficking
drugs or supplying criminal gangs with automatic weapons? In the current conflict
across the border, the drug cartels are armed with a much more powerful arsenal than
the local police. Law enforcement officials confront similar challenges when
combating the growth in cyber crime. Many criminals use the defense of, “I just did it
to see if I could get into the system and didn’t know what I was getting.” Being a red,
white, grey, or black hat may depend on a person’s perspective. Many of these actors
see themselves as beneficial to the network security industry and downplay their
influence on cyber criminals (which they often once were). Movies, such as “Hackers”
continue to glorify the actions of teen-agers who break and enter systems in cyberspace,
when such actions against a physical facility would be clearly viewed as trespassing.
As stated above, as recently as 2004, hacking was not even considered a criminal
offense in Brazil.
There may be several more similarities, but is should be evident from this
presentation that there are many conceptual similarities between these two types of
non-traditional threats to national security. With this in mind, I will discuss how
borders have played a role in the international war on drugs to determine if such
measures can illuminate the complexities of countering attacks through cyberspace. As
stated, in this age of globalization, it is virtually impossible to detect contraband
crossing national borders. The US has, for example, 106,000 miles of physical borders
and shorelines and over 400 million people transit those borders yearly [26]. Though
we cannot completely secure the borders against drug smuggling, they still seem to
play an important role in efforts to combat the trade. The recent measures by both the
US and Mexican governments along their shared borders highlight the political
importance of actions taken to secure borders against the movement of drugs. Arguable
the measures were enacted due to a perceived loss of control of the borders to the drug
cartels. Peter Andreas, Harvard Professor and the author of Border Games [29], asserts
that border control measures are an important symbolic and perceptual response that
the state is defending its sovereignty and its citizens from an existential threat. By
“sending in the troops,” the state can demonstrate its moral resolve and commitment to
maintaining its territorial integrity. Even if there is little empirical evidence that any
measures enacted to defend the borders against the flow of drugs has an effect at
reducing the inflow to the US, there is tremendous pressure to take action. Besides
demonstrating resolve, the visible actions remove pressure to confront the more
difficult but root causes of the drug trade-the insatiable demand.
In addition to the largely symbolic nature of recent actions on the US-Mexican
border, law enforcement officials do attempt to achieve a deterrent effect with their
actions. In this and other border regions, measures have gone beyond the dedication of
more personnel. Enforcement measures have relied on improved surveillance
technology but also such measures as “pushing out” borders [19]. For example, in both
the case of European Union countries and the United States, the immediate neighbor
countries are enlisted to “thicken” the border defenses. In the case of the United States,
the problem of drug trafficking is not limited to the immediate border. The drugs
originate in several source regions and many are funneled through Mexico. According
to Andreas [19], by supporting Mexican efforts deep in Mexican territory, a larger
buffer zone is created while supporting the smooth flow of legal cross-border
commerce. In the European Union, member countries encourage neighboring countries
to improve coordination with their law enforcement efforts by making tighter law
enforcement actions pre-conditions for admission to the EU. Lastly, efforts can also
focus on commercial trans-shippers of legitimate goods who depend on speedy transit
of international customs facilities. Flynn [26] suggests that it is in the interest of
transnational shipping companies to tighten their own logistics and transportation
procedures. As the logistics infrastructure continues to improve and widen the markets
for perishable goods and “just-in-time” deliveries, shippers are under increasing
pressure to maintain delivery schedules. Therefore, they have a tremendous incentive to
avoid any potential delays that could be created if they are found to be lacking controls
on their cargo. Customs officials could use this incentive structure to their advantage
and encourage commercial trans-shippers to help reduce the potential smuggling of
illicit drugs.
These three example measures could have implications for patrolling the borders in
cyberspace. First of all, as threats in cyberspace become increasingly securitized, we
can expect the same pressure for national governments to take action as they have done
in the wars on drugs. This will undoubtedly entail largely symbolic actions to attempt
to secure national borders in the domain. In the case of the US, such efforts may be cast
as an attempt to “regain the control of cyberspace” it ostensibly maintained during the
early years of Internet development. At the time, it was managed by the US
Department of Defense and then the National Science Foundation. The symbolic
gestures to “regain control” can be reified by technological border control points,
attempting to thicken the cyber borders, or both.
For example, a border control point could be established at the terminus between
undersea cables and fiber optic lines. At these points, customs, law enforcement, or
other agents of the federal government could employ any of several technical solutions
such as deep packet inspection devices or Anagran flow management devices [17,30].
Other solutions suggest labeling traffic to identify countries of origin and destination
[31]. The intent here is not to debate the technical or practical feasibilities of such
measures 8 . Without employing any such measures, there is no empirical evidence
available to determine their efficacy, or if they will slow Internet traffic appreciably.
The point here is to show that several measures have been researched and, enacting any
or all would, at a minimum, be symbolic statements to assert sovereignty over national
territory in cyberspace.
More practical measures would mirror the defense-in-depth approach taken by
Europe and the United States to combat drug trafficking and other CTA activities. For
example, nations with more developed legal and law enforcement regimes could
encourage neighboring nations to improve their legal regimes. Developed nations could
also provide technical support to others’ national cyber security centers. One unique
characterization of cyberspace is that neighboring nations in the domain are not always
8
For an in-depth discussion of a multi-agency Internet Border Inspection Station concept, see Upton 2003.
For details on the Anagran technology, see http://www.anagran.com/products_fr_1000_intelligent.php.
physically contiguous. However, that should not limit the possibilities for cooperation.
As with drug trafficking, the focus must be both on nations where attacks have
historically transited, and those where the attacks are perceived to be originating. A
recent effort in Europe to “thicken the cyber borders” has been the broad adoption of
the Council of Europe Convention on Cyber Crime. Six countries signed, ratified and
entered it into force by 2004. However, since that time, an additional nineteen countries,
to include the United States have adopted the convention9.
Additional defense-in-depth measures could focus on the cyberspace common
carriers- the Internet Service Providers and Backbone companies. They are the carriers
of the legitimate traffic in which the contraband is hidden. Like the international trans-
shippers of physical goods, these are the commercial interests that would be adversely
impacted by tightened border controls that may result from the emplacement of
government-monitored border inspection devices. Employing such a suite of inspection
tools, which would adequately provide for the protection of civil liberties, would
invariably slow Internet traffic. Therefore, ISPs would be expected to have a great
incentive to support the improvement of self-regulated inspection regimes. If they can
be motivated to improve their internal procedures to help law enforcement combat
cyber attacks, then there will be less pressure for more restrictive national-level.
Perhaps, the absence of a real threat of employing federal border security measures has
contributed to neglect on the part of ISPs to better control the activities of their
customers. Regardless the exact point of entry of goods and people in any domain,
states have sovereign rights over all their territory and can also pursue legal recourse
against cyber crimes committed anywhere within their borders. Though effectiveness
has been limited, we must continue to rely on state responses for the foreseeable future.
This section of the paper used a comparative case study approach to identify
lessons that could be taken from the fight against international drug trafficking and
applied to cyber security. The next section of the paper will take a fundamentally
different approach and explore the use of a game-theoretic construct and novel
quantitative methodology to address the issue. The analysis expands on a theory that
has been previous employed to research situations in which the security of one actor
depends substantially on the actions of other actors in their system.
3. Interdependent Security Theory
In his book, Micromotives and Macrobehavior, Nobel laureate Thomas Schelling [32]
described the concept of binary, “either-or,” choices that create externalities on the
decisions of others. To explain the concept, he described several different situations
where the question was not about how much anyone does, but how many make one or
the other choice. For example, the decision to follow daylight savings time or joining a
boycott would be considered binary. The interesting implication of Schelling’s model
is the potential to “tip’ the collection of decision makers from one decision to the other.
This tipping effect could reduce the potential social costs when not enough actors
initially make the socially beneficial choice. The model can also be applied in
situations where actors must coordinate security decisions. Economists Kunreuther and
Heal [33] built on this concept of interdependent decision-making after the events of
9
Assessment based on a table from the Council of Europe website at:
http://conventions.coe.int/Treaty/Commun/ChercheSig.asp?NT=185&CM=&DF=&CL=ENG
9/11. They introduced a game-theoretic approach to explore more deterministic

outcomes of a class of binary choices, the interdependent security (IDS) investment
decision. Kunreuther and Heal’s focus has been on the existence of Nash-equilibria in
the decision whether or not to invest in security measures. Of particular note, in an IDS
model, a fundamental assumption is that an indirect attack, or an attack that originates
from within the system due to a failure by a partner actor, cannot be defended against.
For example, no matter how much an airline invests in security at their terminal
location, if a partner airline allows a bag with a bomb to be transferred to their airplane
bypassing the terminal inspection, the investment is for naught.
Arguably, this analytic framework can be applied to the cyber security problem at
the nation-state level. In this case, the actors making a security investment decision
would be national level governments. The investment could be undertaken by private
or public agents, but the security action would be those measures required to secure the
critical infrastructure cyber systems and the information systems containing data of a
national security value, in other words, the targets of the attacks specified early in the
paper. Attackers could be any actor that has the ability to hold the above targets at risk
either through a direct attack on a country from a location therein, or indirectly through
other countries’ national IT systems.10 In other words, attribution of an attack is not
necessary to use this framework.
Understanding that inter-connectivity and inter-dependence is a complex issue at
the national level, this analysis must generalize these aspects of the model. Specifically,
I make several important assumptions. First, I must assume the state, or an agent acting
on its behalf (such as the Department of Homeland Security in the US), can maintain
some level of control over the actions of the owners and operators of the national
critical infrastructure. In other words, the national government must have the ability to
ensure a cyber security investment is executed. Secondly, the analysis assumes that it is
possible to invest at a level that significantly protects these systems from failure or
information theft that would create an existential risk to the nation. In other words,
applying the theory at a national level means that not every single attack must be
thwarted as national systems do have some resilience. Lastly, it assumes that actions
taken to secure a nation’s critical assets in cyberspace can be made visible to others 11.
Even with the above assumptions, there are two additional challenges to applying
the IDS theory in this situation. First, it is difficult, if not impossible, to obtain the
empirical data regarding the security investment decisions of these agents (especially if
many nations do not oversee cyber security at the national level), and the
interdependent nature of the decisions by so many actors can be difficult to calculate as
their decisions are updated. In such situations, an agent-based model can be useful.
Agent-based models employ object-oriented software programs to model the behaviors
of inter-acting agents endowed with specific parameters that govern their behavior.
Such tools can also model the dynamics created by changes in the behavior of the
agents. They are ideal for game theoretic decision problems amongst many actors. In
order to conduct this analysis using an agent-based modeling technique, the first
decision that must be made is the size and composition of the sample. Who are the
representative agents making the investment decisions? Conceivably, I could have
10
In other words, the exact identity or sponsor of the hostile act is not required for this model to function.
11
For such an assumption to be valid, it would imply close coordination of efforts amongst those who have
chosen to make the necessary investments. For example, partner nations could have liaisons working in each
other’s national response centers.
constructed a model of all the nations of the world. It would not have created a
computational problem for an agent-based, but it would probably not closely reflect the
“real world” in cyberspace12. Instead, I identified a more practical sample based on the
most highly connected nations in the world. The sample network used for this analysis
is comprised of the 21 largest countries, in measure of 2008 bandwidth connectivity, as
reported by Telegeography13. While this sample does not necessarily represent existing
cyber security cooperation, it is intended to capture two features. First, nations with the
most extensive international Internet connections can be assumed to be those for which
cyber security is most critical. Secondly, it identifies which countries have the greatest
imperative to work together due to their existing high level of inter-connectivity. A
graphic depiction of the sample network is presented in Figure 1. Again, this is only a
representative sample to illustrate the potential for this type of analysis. The nodes are
representatives of the 21 countries with the highest amount of Internet traffic, and they
are connected by non-valued links symbolizing the overall traffic route (i.e., there may
be several physical connections between the nodes).
The particular structure of this network has a potential benefit for most of the
involved nations. For example, assuming that the vast majority of international traffic
Figure 1. Sample Network of Nations
12
For example, the investment decisions of the Maldives may have little effect on those of the United States.
13
http://www.telegeography.com/products/map_internet/index.php
must flow through one of the few central actors in this network, a cyber attack that
successfully breaches the critical systems in any nation must, in many cases, flow
through (or originate) in one of the central countries in order to reach one of the other
nations connected to the world only through that central nation. Though this does not
reduce the external risk for the few central actors, the structure greatly reduces the
externality problem encountered by the rest of the countries in this network. Therefore
in practice, though the sample contains 21 agents (nation-states), the interdependent
nature of the security investment decision for most “non-central” nations is reduced
immediately to a two-player game minimizing risk estimation14.
Constructing an agent-based program for this model is fairly straightforward.
However, an accurate calibration can be challenging. A complete discussion of the
construction and calibration is contained in Appendix 1. Based on the data collected
from several cyber security sources, the model was run at varying levels of externally
generated risks by altering the probability of an indirect attack. The output of the model
is most easily interpreted by comparing graphic depictions of the investment decisions
over time. Encouragingly, the series of graphics in Figures 2-4 suggest that the agents
in this sample network, the highly connected nations, could behave in a manner
consistent with the IDS theory.
For the model run depicted in Figure 2, the probability of an indirect attack is set at
0.2. In this instance, the probability is sufficiently low that all the agents in the model
find it feasible to invest in security to thwart direct attacks from “outside the system.”
Skipping Figure 3 for the moment and moving to Figure 4, we find the opposite results.
In this case, only a handful of minimally connected nations choose to invest. Since no
other nations choose to join these actors once they have decided to invest, the system
contains both investors and non-investors at equilibrium. The more interesting result is
obtained when the risk of indirect attack is at a point in between. In Figure 3, the
probability of indirect attack is set at 0.3. In this scenario, all but the central actors find
Figure 2.
Full investing
on first round
14
Interestingly, the sample network does exhibit features that make it theoretically feasible to extend these
findings to a much larger set of nations. The network appears to be a scale-free form. For a good discussion
of the implications of this characteristic, see Barabasi [34].
Figure 3.
Full
investing after
several rounds
Figure 4.
Mixed
investing at
equilibrium
it advantageous to invest regardless the decision made by other agents. After T 2, the
risk of indirect attack to the central actors has now been reduced to the same level faced
by the periphery in T1. As a result, the central actors now find it in their interest to
invest. In other words, the conditions are such that the system will cascade to a state of
full investing as predicted by the IDS game-theoretic model.
These results demonstrate how the actions of one country to improve their national
cyber security, whether enacted publicly, privately, or in concert, could positively
impact the decisions of other countries confronted with similar challenges. Though we
are far from such a reality, the results suggest that when it is possible for a nation to
secure its territory within cyberspace, then borders become increasingly important
components of cyber security. Namely, they become important as concepts that
delineate which portions of cyberspace may be more secure than others. There may not
be a cyber security measure enacted directly at physical border post, but when state
actors can determine at which point their activities stop and the actions, or inactions, of
a neighbor occur, than they can better determine the relative values of enacting their
own security measures.
Hopefully, this model demonstrates the importance of coalition building. The
border that may be important may not be national borders, but the borders between the
investing coalition and others. Until the entire system becomes part of the coalition,
there will always be a border between those inside the security umbrella and those
outside. As the security umbrella grows and strengthens, the borders will be more
sharply delineated. In addition, those outside the security framework may increasingly
become the focus of attacks themselves, or the assumed source of malicious actions.
Neither condition would be considered very favorable and would lead to increasing
pressure to join the coalition of those who have chosen to secure their territory in the
domain.
4. Conclusions
This article sought to demonstrate the relevance of borders in issues regarding cyber
security. As stated earlier, nation-states have borders. Domains, as merely mediums in
which we interact, do not have borders. However, the relevance of a nation’s borders in
each domain is related to a nation’s willingness and ability to assert their sovereignty in
them. As long as threats are directed at nation-states, and legitimate response actions
are retained by the state, they will remain important actors and their borders will
continue to be relevant. Borders can be equally important in cyberspace because
borders define boundaries of sovereignty regardless the domain and the ability to locate
them physically. Even though borders have become less significant for all legal
commerce, they have become even more significant for policing action against trans-
national threats. To explore their significance, I employed two dissimilar frameworks
in an effort to broaden the discussion beyond purely technical and legal dimensions.
In comparing international drug trafficking with the national security elements of
cybercrime, we see that there are several similarities. These similarities suggest that we
can learn lessons from measures to secure borders against the shipment of illicit drugs.
Whether the measures are largely symbolic or actually effective, they demonstrate
national resolve and a determination to exert sovereignty. If the current drug wars teach
us anything about national responses to transnational problems, the borders will
become important in the fight to secure cyberspace if only for their political
significance. More effective measures seem to center less on a tight control of the
border itself, and more on improving the behavior of companies engaged in legitimate
trade across the borders, and the behavior of surrounding countries within their
territories.
The agent-based model of the IDS problem provided a theoretically different
perspective. In this model, the exact location of the borders is not relevant. What is
more important is the potential of, and benefit from coalition-building by actors
responsible for securing their portions of cyberspace. If the IDS model teaches us
anything, it is that we must work together on an international level. Unilateral efforts to
secure borders are a losing proposition in today’s interdependent reality. Nowhere is
this interdependence more visible than in cyberspace.
The paper also included some considerations for public policy. Specifically, efforts
can be made to induce better self-regulation of ISPs to avoid more intrusive border
control measures. Also, measures to increase the visibility of national and private
security measures can increase the incentives for others to make similar investments,
and reduce the potential for a security dilemma arising between nation-states in
cyberspace. Continuing to support and encourage neighboring countries to improve
their legal regimes and law enforcement efforts is also an important step. However,
some findings lead to further policy questions. For example, how do we change cultural
attitudes to criminalize hacking behavior? Also, how and when do we increase
visibility and share information with countries that may be the source of many threats
in the domain? A final note of caution for policy consideration: sovereign powers must
be careful not to use the concepts of borders to curtail the progress our nations have
made to connect and better the world via this evolving and expanding environment. In
other words, we can no more lock down the borders to counter malicious actors in
cyberspace than we can lock down our nation’s physical borders to fight terrorists and
drug-traffickers.
If we accept that nations can play a role to improve cyber security in their country
(within their borders) and influence others to do so as well, then there will continue to
be an important role for borders as a physical and legal concept. However, if we find
that it is not plausible for the state to affect security in its portion of cyberspace either
technologically or conceptually, then the existence of borders in any sense becomes
less relevant. Assuming this paper successfully demonstrated the former case is more
true than the later, then regardless their exact physical location, the very existence of
borders demonstrates a need for us to work together as an international community to
develop transnational solutions.
Appendix 1: Agent-based Model Construction and Calibration
Most agent-based models are constructed in an object-oriented computer programming

language and they interact in an environment. As described in the paper, the agents in
this analysis interact within a network. Based on the IDS model equations, the
significant parameters are as follows:
Agent Parameters
• c = Cost of investing in security to a level that defeats existential threats
through cyberspace
• L = Loss of critical information from a successful attack
• pii = Probability of a direct, successful attack
• pji = Probability of an indirect, successful attack that occurs from within the
network of highly-connected nations
• Network neighborhood
• Investment State (invest or not invest)
The next challenge is estimating the values of these parameters. All of the nations
in the sample have made some level of investment, and the cost, at a national level of
achieving an efficiently secure state of security can only be guessed in the absence of
specific data on the threats to each. Therefore, another simple convention is employed.
The agents are heterogeneous in that initial endowment of c, L, and p ii, are randomly
distributed amongst the agents. However, the possible values of these variables are
normally distributed about a mean value. This convention allows us to assess the
actions of the agents when changing the probability of indirect attack, p ji, while holding
other parameters within realistic bounds 15 . The value for pji for each agent is also
normally distributed about a mean value, but it is a single variable for the agent’s entire
network neighborhood16. After consultation with cyber security experts regarding the
potential costs and losses at the firm level 17, the remaining parameters were estimated
by the following mean values:
c = $1,000,000
L =$10,000,000
pii = 0.4
Though the empirical data is not available, a mean value for pii was set at 0.4 based
on a recent SANS Institute report [35] regarding attacks on firms in several industries.
While holding these parameters constant, pji was varied from 0 to 1 to explore whether
this sample network can behave in a manner predicted from IDS theory.
In addition to the agent parameters, there are rules that govern agent interaction.
The agents make the security investment decision according to the interdependent
security pay-off algorithms of the Kunreuther and Heal model18. In the current model,
the behavior of the agents is determined by the following:
Interaction Rules
• Identify how many others in your network neighborhood have not yet invested
in security
• Calculate the external risk created by their decision not to invest
• Determine if the external risk is less than or greater than the cost to invest
• If the cost is less, then decide to invest. If not, then decide not to invest
• Once all agents have made this decision, everyone changes their state as
appropriate
• Repeat the above process until no one wants to change their state again
Initially, the agents are in a state where they have not invested. Since Kunreuther
and Heal (2003) identified that the cost of the risk externality as the significant limiting
condition, agents choose to invest in security when the inequality, c < p ii(L - X), is true.
In this equation, X is the externality generated by others in the network that have not
invested in security.
References
[1] B. Buzan, O. Wver, J.D. Wilde, O. Waever, Security: A New Framework for
Analysis, Lynne Rienner Pub, 1997.
15
In this assessment, ‘realistic bounds’ means relative to the other agents. The author did not attempt to
estimate the actual, absolute costs of any of the parameters nor was it necessary to do so.
16
In other words, the risk from each other agent, to which an agent is connected, is the same. This
convention was necessary to generate a less complex decision algorithm for the basic model.
17
For this analysis, it is only important that the relative values be appropriate and therefore it is assume that
the values at the firm level can be scaled to the national level with the understanding that, just like the
difference between small and large firms, the costs at a national level can vary widely as well.
18
See Heal and Kunreuther [36] for a complete explanation of the pay-off matrix according to a two-person
game.
[2] J. Lewis, Securing Cyberspace for the 44th Presidency, Center for Strategic and
International Studies, Washington, D.C., 2008.
[3] Office of the Press Secretary, (2009).
[4] B. Kahin, C. Nesson, Borders in Cyberspace: Information Policy and the Global
Information Infrastructure, The MIT Press, 1997.
[5] C.C. Joyner, C. Lotrionte, Eur J Int Law 12 (2001) 825-865.
[6] A. Wolfers, Discord and Collaboration: Essays on International Politics, The
Johns Hopkins University Press, 1965.
[7] A. Streltsov, Diarmament Forum 3 (2007) 5-14.
[8] J.H. Herz, World Politics 2 (1950) 157-180.
[9] B. Buzan, People, States, and Fear: The National Security Problem in
International Relations, University of North Carolina Press, 1983.
[10] D. Denning, in:, J. Arquilla (Ed.), Networks and Netwars: The Future of Terror,
Crime, and Militancy, Rand, Santa Monica, Ca, 2001, pp. pgs 239-288.
[11] S.P. Gorman, in:, P.E. Auerswald, L.M. Branscomb, T.M.L. Porte, E.O. Michel-
Kerjan (Eds.), Seeds of Disaster, Roots of Response: How Private Action Can
Reduce Public Vulnerability, 1st ed., Cambridge University Press, 2006, pp. 239-
257.
[12] F. Kramer, S. Starr, L. Wentz, eds., Cyberpower and National Security, Potomac
Books, Dulles, Virginia, 2009.
[13] C. Wilson, Information Warfare and Cyberwar, Congressional Research Service,
The Library of Congress, 2004.
[14] DHS, National Infrastructure Protection Plan, Department of Homeland Security,
2006.
[15] S. Gorman, A. Cole, Y. Dreazen, Wall Street Journal (2009) 3.
[16] Energetics, Roadmap to Secure Control Systems in the Energy Sector, 2006.
[17] O. Upton, Asserting National Sovereignty in Cyberspace: The Case for Internet
Border Inspection, Master of Arts Thesis, Naval Post Graduate School, 2003.
[18] L. Shelley, Geo. J. Int'l Aff. 6 (2005) 5.
[19] P. Andreas, International Security 28 (2003) 78-111.
[20] J. Leyden, The Register (2009).
[21] T. Gibb, Bbc (2004).
[22] T. Thomas, Cyber Silhouettes, 1st ed., Foreign Military Studies Office, Fort
Leavenworth, 2005.
[23] R. Deibert, R. Rohozinski, Tracking Ghostnet, Centre for International Studies,
University of Toronto, Toronto, 2009.
[24] S. Flynn, in:, Beyond Sovereignty: Issues for a Global Agenda, Palgrave
Macmillan, 1999, pp. 44-66.
[25] B. Krebs, Report: Russian Hacker Forums Fueled Georgia Cyber Attacks (2008).
[26] S. Flynn, Foreign Affairs 79 (2000) 57.
[27] W.S. Baer, A. Parkinson, Security & Privacy, IEEE 5 (2007) 50-56.
[28] H. Lafranchi, Christian Science Monitor (2009).
[29] P. Andreas, Border Games: Policing the U.S.-Mexico Divide (Cornell Studies in
Political Econ, Cornell University Press, 2001.
[30] P. Bartram, (2009).
[31] J.E. Molini, Computers & Security 16 (1997) 189.
[32] T.C. Schelling, Micromotives and Macrobehavior, New, Norton, New York,
1978.
[33] H. Kunreuther, G. Heal, Risk and Uncertainty 26 (2003) 18.
[34] A. Barabasi, Scientific American 288 (2003) 9.

[35] SANS Institute, Top Ten Cyber Security Menaces for 2008, SANS Institute,
2008.
[36] G. Heal, H. Kunreuther, Journal of Conflict Resolution 49 (2005) 201-217.
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-106
Towards a Global Regime for Cyber

Warfare
Dr Rex HUGHES
Cyber Security Project, Chatham House, London
Abstract. With two years having passed since the infamous cyber conflict
between Estonia and Russia, international society still lacks a coherent set of
principles, rules, and norms governing state security and military operations in
cyberspace. For parties committed to promoting the cause of peace and stability in
a multipolar world, this is a troubling notion since history shows that the
likelihood of a new arms race is high when disruptive technologies dramatically
alter the means and methods of war. As more nations aspire to project national
power in cyberspace, a new digital arms race appears to be imminent if not already
upon us. Thus, there is a central question confronting international society and
more specifically the diplomatic community in cyberspace: What steps can be
taken both today and into the future to forestall a major arms race and interstate
competition in cyberspace? In order to begin addressing this complex question
from the perspective of the Euro-Atlantic Community, this paper discusses both
the challenges and opportunities of regulating 21st century cyber warfare. The
paper is divided into three sections. Section 1 examines the evolution of the laws
of armed conflict (LOAC) since the late 19th century. Section 2 examines how the
LOAC apply to cyber warfare as viewed primarily from a US perspective (since
US scholars have dominated the international regime discourse thus far). Section 3
examines what is needed to create a global regime for cyber warfare and
specifically the role that NATO and the Euro-Atlantic Community can play.
Keywords. Law of Armed Conflicts, Cyber Warfare, Revolution in Military

Affairs, Global Regime
Because the entire law of war regime has been built upon a Westphalian foundation,
the transformative properties of cyber warfare are just as breathtaking. We are left
pondering some fundamental questions - what constitutes force? What is a hostile act?
When is self-defense justified in response to a cyber attack? Is the use of traditional
means of force ever justified in response to a cyber attack? These are not easy
questions and the international legal regime is lagging far behind the problems
presented by the increasingly sophisticated technological possibilities in this area.
--Lt. Col Jeffrey K. Walker[1]
Introduction
When viewed systemically, the current generation of cyber weaponry demonstrates an

enormous potential to alter the means of hostile attack and in turn of response. While
our 21st century armed services are adjusting to the revolution in military affairs
(RMA), the broader community of business, transportation, energy, research, health,
academic, and social services look to their national leaders to provide plans and to
R. Hughes / Towards a Global Regime for Cyber Warfare 107
conduct operations that will protect their domain of cyber space. Cyber defense for
those old enough to remember may call to mind the home front nuclear alert drills plus
the bunkers or bomb shelters constructed in the post WWII decades. In cyberspace both
military and civilian networks are potential targets.
Overarching questions confront us: What is the current state of cyber warfare when
viewed from an international affairs perspective? What options are available to policy
makers that seek to fashion a global regime to govern 21st century cyber warfare? And
more specifically to the theme of the first NATO CCD COE cyber war conference,
what role can an international military alliance such as NATO play in advancing such a
regime?
Since the enormous attack on Estonian digital networks, governments around the
world have ordered their respective military branches to develop new offensive and
defensive cyber capabilities. Some states have even gone as far as to create national
cyber command authorities, as is evident in the United States [2]. However, as the
attacks mount and more advanced ‘cyber weapons’ are introduced to the digital
battlefield, there is little certainty or international consensus on the rules, or lack
thereof, for governing modern cyber battles or larger warfare. Air Force Gen. Kevin P.
Chilton, the head of U.S. Strategic Command (STRATCOM) issued a statement in
May of this year that ‘The Law of Armed Conflict will apply to this domain’[3].
STRATCOM defends the Pentagon’s Global Information Grid at home and abroad
through its Strategic Command Joint Task Force-Global Network Operations (JFT-
GNO). Attempted penetrations of public and private systems number in the tens of
thousands a day. As a commander who provides information for decisions by the US
President and the Secretary of Defense, Gen. Chilton said that all combat options
should be on the table for a US response to a cyber attack. He noted that many attacks
thus far have been for the purpose of espionage, and that there can be an argument
about the ‘semantics of attack versus espionage and intrusion’ [4].
1. REGULATING WARFARE
The Principles and Norms of Armed Conflict
The law of armed conflict (LOAC also commonly referred to as the ‘laws of war’ or jus
in bello) as understood today originated in the mid-19th century, as did the
humanitarian regulation of conflict and violence.1 Since their early beginnings these
laws applied primarily to interstate conflict as carried out by uniformed armed forces
between two or more states. The principles, rules, and norms that guide today’s LOAC
can be found in a variety of sources: customary law, international treaties, judicial
decisions, legal philosophers, and military manuals. Although the customs of the
LOAC can be traced as far back to the 15th century medieval Europe, its more modern
origins date back to the American Civil War of 1861-1865 [5]. Until that era Dale
Stephens and Michael Lewis, note that ‘there was no meaningful jus ad bellum because
the right to resort to force was essentially unchallenged’. The ideals of knighthood in
the Middle Ages provided some restraint against certain warfare cruelties [6]. Not until
the 17th century was there a systematic legal code on war and peace. The Hugo Grotius
1
Sections of the LOAC that deals explicitly with civilians are commonly referred to as international
humanitarian law.
108 R. Hughes / Towards a Global Regime for Cyber Warfare
work of 1625, On the Law of War and Peace (De jure belli ac pacis), was based on the
natural law. In the 18th century natural law and the Golden Rule were formulated by
Emerich de Vattel in his 1758 work, The Law of Nations (Droit des gens). It was not
until the American Civil War that the laws of armed conflict were codified and adopted
by the leading world states. The first Geneva Convention was agreed upon in 1864 and
employed the US Lieber Code (US War Department, General Orders No. 100, 24 April
1863) as a baseline. By 1868 in St. Petersburg a treaty, as noted below, was signed by
leading nations/empires (excluding the US) that concerned regulating warfare methods
and means.
Regulation of war and violence under humanitarian principles arose in 1863 with
the establishment of the International Committee of the Red Cross (ICRC). During the
1876-1878 Russo-Turkish War the Ottoman Empire established the Red Crescent and
there followed mutual agreement for respect from both sides whether honoring the
principles of the Red Cross or the Red Crescent. The ICRC acts as a guardian and
promoter of international humanitarian law.2 The establishment of the First Convention
of the Geneva Conventions in 1863 followed the formation of the ICRC and concerned
the welfare of the wounded, civilians, shipwrecked, and prisoners of war. Four
Conventions were adopted/revised through 1949. Amendment protocols about victim
protections were added from 1977 through 2007. In 1899 and in 1907 peace
conferences were held in The Hague for the purpose of regulating weapons in war and
the customs and laws of war. The Geneva Protocol to the Hague Convention of 1925
prohibited other uses of gas and biological weapons. Later treaties of 1972 and 1993
covered the production, storage, or transfer of these weapons. A 1951 UN Convention
Relating to the Status of Refugees covered post-WWII European refugees and was
expanded in 1967 to cover other refugees without time or geographical limitation.
While also referred to as the ‘Geneva Convention’ this UN treaty is actually not part of
the Four Geneva Conventions [7].
New Weapons and LOAC
It was not until the 1868 ratification of the St. Petersburg Declaration that emerging
military technologies were subject to any type of international legal review. This action
is officially stated as the Declaration Renouncing the Use, in Time of War, of
Explosive Projectiles under 400 Grammes Weight. A rash of new weapons introduced
during the American Civil War, including armoured warships, submarines, as well as
land mines, machine guns, projectiles filled with clear glass or explosives, exploding
bullets, and dumdum bullets designed to flatten on impact. Weapons that caused
unnecessary physical harm prompted delegates at St. Petersburg to issue a statement on
this vexing issue in their declaration:
The Contracting or Acceding Parties reserve to themselves to come hereafter

to an understanding whenever a precise proposition shall be drawn up in view
2
The ICRC explains its mission as ‘an impartial, neutral and independent organization whose exclusively
humanitarian mission is to protect the lives and dignity of victims of war and internal violence and to provide
them with assistance’. The organization ‘endeavors to prevent suffering by promoting and strengthening
humanitarian law and universal humanitarian principles’ [8]. The seven Fundamental Principles of the
International Committee of the Red Cross are: humanity, impartiality, neutrality, independence, voluntary
service, unity and universality. It was not until 1929 with the amendment of Geneva Convention (Article 19)
that the Red Crescent was officially recognized as part of the ICRC.
of future improvements which science may effect in the armament of troops, in

order to maintain the principles which they have established, and to conciliate
the necessities of war with the laws of humanity.3
In 1977, Protocol 1 of the Geneva Convention also cited the need to carry out
international reviews of new weapons. The intent of Article 36 is to determine the
lawfulness of new weapons before they are developed, acquired, or incorporated into a
military’s arsenal. Legal experts have also argued that scrutinizing the legality, means,
and methods of new warfare is so basic that it applies to all nations, even those not a
party to Protocol 1. Thus, states following international law must ensure that new
innovations do not run afoul of international obligations. Article 35 states:
In the study, development, acquisition or adoption of a new weapon, means or

method of warfare, a High Contracting Party is under an obligation to
determine whether its employment would, in some or all circumstances, be
prohibited by this Protocol or by any other rule of international law
applicable to the High Contracting Party.4
Article 6 is complemented by Article 82 of Additional Protocol 1, which requires

that legal advisers be accessible at all times to battle commanders and that ‘on the
appropriate instruction to be given to the armed forces on this subject’. Thus, the basic
aim of both the St. Petersburg Declaration and Protocol 1 of the Geneva Convention is
to ensure that armed forces will carry out battlefield hostilities in strict accordance with
the principles, rules, and norms as established by the LOAC. However, one
problematic area is found in Article 36 that does not delineate specific means by which
legal review of new weapons and methods of warfare will take place, leaving actual
practice open to much interpretation. Few states have actually codified this mandate
into state practice except for the United States and Sweden.5 In 1999 the 27th
conference of the International Conference of the Red Cross and the Red Crescent
encouraged states to, ‘to establish mechanisms and procedures to determine whether
the use of weapons, whether held in their inventories or being procured or developed,
would conform to the obligations binding on them under international humanitarian
law’.6 In 2003, in light of war in the Middle East, the ICRC reaffirmed this goal ‘the
legality of new weapons under international law’ and ‘in light of the rapid
developments of weapons technology and in order to protect civilians from the
indiscriminate effects of weapons and combatants from unnecessary suffering and
prohibited weapons’.7
3
Excerpt taken from the section renouncing the Use, in Time of War, of Explosive Projectiles Under 400
Grammes Weight, St Petersburg, 29 November / 11 December 1868.
4
Article 35, Protoco1, 1977
5
The United States and Sweden both established formal review practices as early as 1974.
6
Section 21, Final Goal 1.5 of the Plan of Action for the years 2000-2003 adopted by the 27th International
Conference of the Red Cross and Red Crescent, Geneva, 31 October to 6 November 1999.
7
Final Goal 2.5 of the Agenda for Humanitarian Action adopted by the 28th International Conference of
the Red Cross and Red Crescent, Geneva, 2-6 December 2003 [hereinafter Agenda for Humanitarian Action].
2. LOAC in Cyberspace
Thus far, there is no international consensus on the application of the laws of armed
conflict to 21st century cyber warfare, likely irregular warfare. This problem stems from
both the loose definition of cyber warfare as well as the lack of precedent through
which to guide present and future law. Another challenge is that not unlike other means
of 21st century warfare, cyber warfare is coming of age in an era where the Westphalian
state order is undergoing vast transformation. As the borderless realm of cyberspace
both ignores and challenges state boundaries, a hands-off policy may disrupt or worsen
cross-border transactions. As explained by Vida Antolin-Jenkins, USN JAGC (United
States Navy-Judge Advocate General Corps), ‘Cyberspace operations for the most part
do not meet the criteria for ‘use of force’ as currently defined by international law.
Defining the parameters of proportional response through analogy is possible, but
creates clear dangers of definitional creep into other areas of international relations that
have long been the subject of long and contentious debate’ (Antolin-Jenkins 2005:
134). Let us remember that Gen. Chilton does not rule out responding with kinetic
force to a cyber attack [2].
Disruptive Technologies on the Battlefield
Throughout history, there are numerous examples of disruptive technologies that have
changed the way war has been waged, often resulting in enormous transfer of powers.
Since hunter-gatherer tribes first wandered the earth, man has sought strategic
advantage through the development and application of new technology. William
Owens (US Admiral-Ret.), applying historian Martin van Creveld’s divisions of
military history, reminds us that the ‘Age of Tools’ (prehistoric age to about 1500 AD)
witnessed battles of the ‘muscular power of men and animals’ using ‘the wheel, the
stirrup and iron weaponry’. In the ‘Age of the Machine’ (18 th thru the 19th centuries)
artillery weapons were eventually wielded by large units of fighting men [9]. The ‘Age
of Systems’ developed between WWI and WWII with radar, long-range aircraft and
radio coordinated ground-to-air attacks. We are presently living in a 21st century ‘Age
of Automation’ where even older munitions and aircraft are deployed by sophisticated
communications technologies as faster and more precise navigation and related digital
devices are currently in development [9].
Using information and/or computer technology to impose one’s will upon an
enemy is a form of warfare-- information warfare (IW). In 1995 Martin Libicki of the
US National Defense University in classifying IW forms stated: ‘Seven forms of
information warfare vie for the position of central metaphor: command-and-control
(C2W), intelligence-based warfare (IBW), electronic warfare (EW), psychological
warfare (PSYW), hacker warfare, economic information warfare (EIW), and
cyberwarfare’ [10]. Robert Hanseman in the late 1990s anticipated how information
warfare will blossom into a full-fledged “Revolution in Military Affairs” (RMA) [11].
Looking specifically to the US global role, Owens sees this RMA as an ‘opportunity to
use the new information technology to change the very nature of our military, in a way
that could reinvigorate American political, diplomatic, and economic leadership in the
world for decades to come’. From a specifically US security perspective, Owens holds
that in this new century the ‘changing world demands a new way of looking at war and
the proper military force’ [9]. This paper maintains that in addition to this needed
military strategy transformation there needs to be a new legal/treaty framework

defining the cyber security regime.
Another question: What is the nature of the force deployed by cyber attackers or
cyber terrorists and is this force an act of war? Michael Schmitt points to an
International Court of Justice (ICJ) finding that ‘supports a conclusion that a use of
force need not be kinetic in nature’ when the ICJ ruled (1986 Nicar. v. US) that
‘although the funding of guerrilla forces was not a use of force, arming and training
them was’ [12]. Today cyber weapons may potentially be in the possession of varied
and mixed-motive warriors, terrorists, or civilians. The legal assessment of cyber attack
or cyber war, Schmitt offers, will come from the community. He concludes that a
‘cyber attack that causes significant human suffering or property damage is obviously
an armed attack justifying a response under the law of self-defence. Schmitt also speaks
to the need for the law to mature since the ‘global community finds itself at the cusp of
normative change’ in seeking a definition of aggression [12]. As this paper urges, the
year 2009 is the time to begin constructing a global cyber security regime to bring
clarity to the place of cyber attacks or cyber terrorism within the 1945 UN Charter.
Info Age Warfare
The transition through the phases of information technology with increasing

dependency upon sophisticated devices and digital applications have led logically to
phases --warfare involving information systems, hence new battlefields or rather,
battlespaces. In one sense, modern information warfare can be traced back to the
introduction of long-distance telecommunications. With the introduction of the
telegraph, telephone, and radio, both civilian and military leaders gained an
unprecedented command and control authority over troops movements and
deployments. With a real-time link established between civilian and military
headquarters, political leadership where possible could exert much more decision
making authority, sometimes to ill effect. US President Abraham Lincoln was the first
commander in chief to use the telegraph to issue orders to his Northern Generals in
real-time. From across the Atlantic, Queen Victoria made extensive use of the telegraph
to communicate her vision to overseas colonial viceroys.
During WWI, communications instruments were fully integrated into land, sea,
and air campaigns. Later the sheer complexity and scale of WWII increased the need
for cybernetic controlled weapons. As firepower increased again throughout land, sea,
and air, the need for more precision and predictability increased. This was especially
true in two areas of artillery and air combat. The Bletchley Park British code-breakers
of WWII decrypted over 3000 daily German Enigma messages through the
computational developments of Alan Turing [13]. The British and US through
technology exchanges perfected their microwave and radar system and eventually were
able to jam the German radars.
The atomic age only increased cybernetic development as the need for even more
precise, reliable, and speedy command-and-control operations increased several orders
of magnitude. The deployment of nuclear weapons required the most sophisticated air
and space communications networks ever developed. Between the 1950s and 1960s,
the space and nuclear race helped propel a host of new information communications
technology (ICT) breakthroughs with the invention of the transistor and the
microprocessor. From the 1970s forward, many of these dual-use military and space
technologies found their way into the civilian sector. Historically, with each generation,
the ability to hunt and destroy has increased.
Peace and Security and the Cyber Challenge
‘A war of aggression is a crime against international peace. Aggression gives rise to

international responsibility’ states the UN Charter (Article 5, Paragraph 2 of the
Definition of Aggression). This statement provides a basis for considering an attack on
the information infrastructure of a nation as an act of aggression. Attacks have the
potential to disrupt a nation’s power grids, transportation links, health care service,
emergency response, financial flows among many other venues. Under this Charter
from 1945, a nation has the right to self defence. Lawmakers, diplomats, and military
strategists need to confront the tasks for defining and framing the regime that will
protect the cyber security for national defence and civil society functions. A major
international military alliance such as NATO has a task suitable to its own mandate or
charters in harmonizing accepted policy on aggression and protective national defence
on the its own membership soil and their protectorates and in its global expanse of
cyberspace.
Antolin-Jenkins sums up the predicament of a digital society: ‘The strategic and
economic power of the increased information awareness and connectivity are coupled
with a hugely increased vulnerability to destruction and attack. This information
network has created a new battleground’. She also reminds her readers that ‘One
estimate is that 95% of military information traffic utilizes civilian networks at some
stage of communication’ [22][14]. How or when could there be a separation of military
from civilian networks? Is an attack on a military network also an attack on civilians,
or visa versa? An advisor to the director of US national intelligence, Steven Chabinsky,
reminds us that even if the laws of war ‘would forbid targeting purely civilian
infrastructure’ we need to consider that ‘terrorists, of course, don’t limit themselves by
the Geneva Conventions’ [15].
Humanitarian Law for Cyber Weapons?
Does cyber warfare fall under the international humanitarian law (IHL)? Jeffrey Kelsey
holds that IHL ‘should evolve to encourage the use of cyber warfare in some situations
and provide states better guidance in the conduct of these attacks’ [16]. He argues that
for a decade or more the ‘potential threat and opportunity of cyber warfare’ have
confronted military planners while the ‘international community has yet to reach
consensus on the application of IHL’. This lack of consensus may be due to a variety of
reasons, from holding that the ’current IHL framework can be applied to cyber warfare
by analogy’ to the realization that vast growth and fluidity of technology would make
potential international agreements obsolete [16]. Kelsey further maintains that ‘IHL
applies to cyber warfare by analogy but contends that IHL must evolve to
accommodate and, in some cases even encourage cyber warfare over conventional
methods’ [16].
The movement in a cyber attack across a neutral state becomes more than a ‘mere
communication signal’, for cyber weapons can cause damage to states as have more
conventional weapons. A weapon the ‘size of a electron’ could be a violation of the
territory of a neutral state according to the Hague Convention that ‘forbids the
movement of weapons’ across a neutral state which risks being drawn into a wider
cyber conflict when its Internet nodes are engaged by a belligerent [16]. What
obligations would a neutral state have as a conduit for cyber attack and mischief?
Kelsey argues against establishing new treaties and in favour of states and their military
commanders to follow established legal principles in cyber combat [16]. The question
remains: What standards should emerge for a cyber security regime under established
peace and war legal principles? Since cyber warfare is still in its infancy, some would
argue that regulating it is a difficult if not an impossible challenge. However, the so-
called catastrophic cyber attack is to be avoided, it would be foolish and impractical
not to establish some type of international rules of the game as deterrence.
3. Towards a 21st Century Global Cyber Regime
Thus far, the diplomatic community has had little to say about the governance of cyber
warfare. Two exceptions of major importance include former diplomats with
knowledge of ICT have in recent months discussed with this author the major
international relations quandaries from cyber threats and attacks plus their own
concerns about diplomatic-level solutions. These former envoys are retired senior US
Amb. Thomas Pickering who served over four decades in major posting for the US
Dept of State, and Amb. David Gross, U.S. Coordinator for International
Communications and Information Policy 2001-08. The latter observes that diplomatic
silence may be attributed largely to a generational gap and a lack of technical
understanding by policy makers since the Internet and associated networks are fairly
recent developments; therefore, cyber security concepts in international affairs are still
a nascent on the part of the diplomatic community (Gross to the author: April 2009).
With wide-ranging diplomatic and corporate experience, Amb. Pickering sees an even
larger problem in that forming an international treaty involves major, prolonged steps
and major questions: What is the problem to be solved? How will the problem evolve
in the future? (Pickering to the author: July 2009). Since cyber warfare is being
conducted and developed during a period of wide interstate trade and general economic
accord or agreement, there is an opportunity to design a governing framework before
an actual global catastrophic attack takes place. Today the questions remains: Can
governments be motivated to take action now before it is too late?
Special regimes have been formed for far-ranging interests or activities, such as
treaties governing the Arctic, Antarctic, canals, international rivers, and outer space.
While somewhat vague or undetermined there appears to be a consensus that outer
space begins where airspace ends [17]. Examples of the treaties governing outer space
include those governing the International Space Station (1998), Registration of Objects
Launched into Outer Space (1975), INTELSAT or International Telecommunications
Satellite Organization (1986), INMARSAT or International Maritime Satellite
Organization (1976), as well as the ITU or International Telecommunications Union
(1932; 1947 as UN agency) [17]. You may ask, ‘Where does cyberspace begin?’ It
would appear that cyberspace begins with the keystroke to log on to a cyber network,
whether from a mega terminal, a PC, a game console, or a mobile telephone.
Eventually, the ultimate venue for cyber warfare governance would be The Hague
as the home of the world’s first Peace Conference and for over a century as the
international centre of justice and arbitration, as well as warfare governance. The
Hague hosts several international organisations, including the UN International Court
of Justice, the Permanent Court of Arbitration, the NATO Consultation, Command and
Control Agency (NC3A). As cyber warfare moves to the forefront of more government
agendas, more questions arise as to how the Law of Armed Conflict and the Geneva
Convention apply to cyberspace. Responses are likely to range on both sides of the
fault line: those who see cyber warfare as fitting neatly under existing LOAC (as well
as under the UN charter), and on the other side, those who see the need for an entirely
new set of international laws and treaties to govern cyber warfare.
A Way Forward
While it is unlikely that these two countervailing diplomatic/legal views will be

reconciled anytime soon, the time is now to begin having this debate in a more serious,
focused manner. Again, because cyber warfare is a complex and dynamic issue, these
debates will need to be hosted in many different venues and viewed from many
different perspectives. NATO is already playing an important role in this debate by
hosting conferences such as this June 2009 Tallinn gathering bringing together the
relevant players from both member states and global partners. For the foreseeable
future, this NATO Center of Excellence can play a critical role in bringing together the
best experts both to analyze and to debate the problems from a number of unique
cultural and disciplinary perspectives.
As the world’s premiere military alliance, NATO is positioned to play a major role
by facilitating significant interstate dialogue between civilian and military planners.
Each year NATO hosts various fora where such engagements take place. The NATO
Global Partnership Program provides a mechanism to reach out to other countries.
NATO could also explore reaching out to other peace and security alliances, such as
ASEAN Regional Forum (ARF) and the Shanghai Cooperation Organization (SCO),
for the purpose of exploring confidence-building measures with that global hemisphere.
Although presently cyber warfare/defense is largely an ungoverned affair, the UN
leadership has already acknowledged the severity of the problem and the need for
governance. UN secretary-general Ban Ki-moon earlier this year announced that the
UN Advisory Board on Disarmament Matters is to include cyber weapons in its arms
list [18]. In his prepared February remarks to this Advisory Board the Sec-Gen stated:
This year you will be considering cyber warfare and its impact on
international security. As you know, there have been many widely reported
breaches of information systems in recent years. With both the public and
private sectors growing increasingly dependent on electronic information,
your work in this area is very timely. It will also complement the efforts of the
panel of governmental experts that will be addressing information security
later this year [19].
The UN’s International Telecommunications Union last year concluded its

agreement with the International Multilateral Partnership Against Cyber-Terrorism
(IMPACT) to conduct the ITU Global Cybersecurity Agenda (GCA) with headquarters
in Cyberjaya, Kuala Lumpur. The GCA seeks international cooperation for
governments, international law enforcement authorities, the private sector, international
organisations, and civil society for the purpose of a secure cyberspace. Through five
areas the GCA if focused on strengthening the legal framework, technical measures,
organizational structure, capacity building, and international cooperation. [20] At its
Cyberjaya headquarters inauguration the ITU-GCA was billed as ‘public-private
initiative’ and a ‘framework for cooperation aimed at finding strategic solutions to

boost confidence and security’ in a networked world [21].
Relevant committees beyond the UN should also begin to debate a proper a
framework for cyber warfare while other international fora can and should play a role.
Significant organisations such as the Institute of Electrical and Electronics Engineers
(IEEE) and the Internet Society (ISOC) are positioned to use their technical legitimacy
and their soft power to press forward with best practices for member states to follow in
securing their own sovereign cyberspace. World Trade Organization (WTO) signatories
should develop an agreement for pledges from nations not to promote or solicit
mercenaries, or attack a member state’s trade infrastructure. On the military side, the
national leading military services should be encouraged to act with transparency where
possible so as not to launch initiatives that unduly contribute to unnecessary cyber arms
race between states.
One of the most difficult governance areas to reconcile will be in the area of police
vs. military involvement in cyber security/defence. As the US-led global war on terror
has shown, there is no clear line of authority when defending against threats where
state involvement is murky at best. Due to the anonymous and secretive nature of cyber
warfare, state involvement is often tricky for producing fool-proof forensics that can
prove state involvement. While each country will ultimately need to sort through this
problem in accordance to its national laws and constitutions, more global debate will be
needed to clarify these issues.
Participants in Tallinn are witnesses to the positive leadership role assumed by
Estonia and leading to the NATO CCD COE here at the scene of the first
acknowledged, major interstate cyber conflict. There is potential for a cyber treaty to
emerge should the North Atlantic Council embrace thoroughly the cyber warfare issue
and engage the NATO Consultation, Control and Command Agency (NC3A) and the
NATO Military Authorities (NMA). The Tallinn cyber convention questions and
discussions are a start, but a protocol or treaty governing the conduct of cyber warfare
needs serious consideration.
4. Conclusion
While cyber warfare is not an entirely new area of modern warfare (at least as viewed
within an Internet world), its current evolution poses many challenges to international
peace and stability. The increasing quantity and quality of online attacks threaten many
parts of civil society that depend on reliable networks and information systems.
Growing evidence of state-sponsored cyber attacks is especially alarming and could
spark a serious arms race in cyberspace. Understandably, a number of countries have
announced plans for full spectrum military cyber commands. As history has
demonstrated, while international law cannot stop states from going to war with one
another, it can go a long ways towards regulating their conduct should hostilities boil
over into actual war. Some may argue that because cyber warfare is still in its formative
stages, it is premature to begin work on a global regime to regulate it. However, it can
also be logically argued that absence of some rules of the game, states will not feel
constrained to develop and deploy cyber weaponry if the consequences are not
understood by both military and civilian planners. While it is difficult to estimate the
true potential for a catastrophic attack to spill over to kinetic warfare between states,
the notion that the threat exists at all is cause enough to begin constructing a regime or
legal framework through which to conduct cyber warfare.
History presents another lesson in that even with the best intentions and resources,
a global cyber security regime will not transpire in short order. It will take many years
to form an effective international consensus that might translate into a revision of the
Law of Armed Conflict as spelled out by the Geneva Conventions. The operative
concept is regime. And, the time to establish a global cyber security regime is now. As
a proper follow-up to the innovative inaugural Tallinn CCD COE conference of 2009,
NATO can and should play an important role by bringing together in short order the
relevant stakeholders to outline a viable cyber security regime. Lt. Col. Walker quoted
in the introductory epigraph to this paper, eight years ago properly challenge the legal
community, and this writer has chosen to extend his challenges to the wider global
policy community.
References
[1] Walker, Jeffrey K. (2001) ‘The demise of the nation-state, the dawn of new paradigm warfare, and a
future for the profession of arms’ Air Force Law Review (51:2001).
[2] Stars and Stripes (2009) ‘Pentagon Steps Up to Fight Cyber War’ (30 May 2009)
http://www.military.com/news/article/pentagon-steps-up-to-fight-cyber-war.html?col=1186032310810
[3] Schogol, Jeff (2009) ‘Official: No options ‘off the table’ for U.S. response to cyber attacks’, Stars and
Stripes (Mideast edition, 08 May 2009).
http://www.stripes.com/article.asp?section=104&article=62555
[4] Gertz, Bill (2009) ‘Cyber warfare plans’ Inside the Ring (04 June 2009)
http://www.gertzfile.com/gertsfile/InsidetheRing.html
[5] Reynolds, Jeffrey. (2005) ‘Collateral Damage on the 21st Century Battlefield’, Air Force Law Review
(56:2005).
[6] Stephens, Dale and Michael Lewis (2005) ‘The law of armed conflict—a contemporary critique’
Melbourne Journal of International Law (6:2005).
[7] ‘Geneva Conventions: A Reference Guide’ (2009) Society of Professional Journalists.
http://genevaconventions.org (Accessed: 01 June 2009)
[8] ICRC-International Committee of the Red Cross (2009) ’The mission’.
http://www.icrc.org/HOME.NSF/
[9] Owens, William (2001) Lifting the Fog of War (New York: Farrar, Straus & Giroux).
[10] Libicki, Martin (1995) ‘What is information warfare?’ ACIS Paper 3: August 1995; National Defense
University Press. http://www.iwar.org.uk/iwar/resources/ndu/infowar/a003cont.html
[11] Hanseman, Capt. Robert G. (1997) ‘The realities and legalities of information warfare’ Air Force Law
Review (42:1997)
[12] Schmitt, Michael N. (2003) ‘The sixteenth Waldmar A. Solf lecture in international law’ Military Law
Review (176:2003).
[13] Keizer, Gregg (2006) ‘British Museum unveils WWII computer replica’ InformationWeek (08
September 2006).
http://www.informationweek.com/news/security/government/showArticle.jhtml?articleID=192700296
[14] Antolin-Jenkins, Cdr.Vida M. (2005) ‘Defining the parameters of cyberwar operations: Looking for law
in all the wrong places?’ Naval Law Review (51:2005).
[15] Tennant, Don (2009) ‘The fog of cyber war’ Computerworld, 27 April 2009.
[16] Kelsey, Jeffrey T.G. (2008) ‘Hacking into international humanitarian law: The prince-les of distinction
and neutrality in the age of cyber warfare’ Michigan Law Review (106:2008)
[17] Aust, Anthony (2005) Handbook of International Law (Cambridge: Cambridge University Press).
[18] Marks, Paul (2009) ‘Pentagon readies its cyberwar defences’ New Scientist (16 arch 2009).
[19] Ki-moon, Ban (2009) ‘Secretary-General's remarks to the Advisory Board on Disarmament Matter’
(New York: 18 February 2009). http://www.un.org/apps/sg/sgstats.asp?nid=3717
http://www.un.org/apps/dsg/sgstatsarchive.asp
[20] ITU-International Telecommunications Union (2008) ‘ITU’s global cybersecurity agenda housed in
Malaysia’ (04 September 2008). http://www.itu.int/newsroom/press_releases/2008/27.html
[21] ITU-International Telecommunications Union (2009) ‘ITU’s global cybersecurity agenda housed in
new centre in Malaysia: IMPACT headquarters in cyberjaya to focus on strengthening network
security’ (20 March 2009). http://www.itu.int/newsroom/press_releases/2009/08.html
[22] Knecht, Ronald and Ronald Grove (2001) The Information Warfare Challenges of a National
Information Infrastructure’
http://web.archive.og/web/2001110717440/http://invorwar.com/mil_c4i//iwchall.hyml-ssim
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-118
What Analogies Can Tell Us About the

Future of Cybersecurity
David SULEKa, and Ned MORANb
a
Principal, Booz Allen Hamilton
b
Senior Consultant, Booz Allen Hamilton
Abstract. For more than a decade, leading experts in government and industry
have warned of an impending Cyber Pearl Harbor, a surprise electronic attack with
the potential to neutralize U.S. military power and cause massive disruptions in
U.S. and global computer networks. This is a powerful historical analogy—but is it
the right one? This paper articulates a framework to better explore and examine the
use of historical analogies in their application to conflict in cyberspace. The
resulting analysis does not seek to argue the Pearl Harbor analogy is a bad one.
Quite to the contrary—our thesis is that while a cyber Pearl Harbor remains a
possibility, is should not be treated by decision makers as an inevitability and that
there may be equally powerful historical analogies to guide future cyber strategies.
Keywords: cyber conflict, cybersecurity, decision science, decision-making,

historical analogies, public policy
Introduction
In their study Thinking in Time: The Uses of History for Decision-Makers, authors
Richard Neustadt and Ernest May speak to the power and perils of making decisions
through the use of historical analogies. They argue for a structured, critical inquiry to
address an issue or crisis rather than leaping to a single analogy for which to formulate
strategies and policy options (e.g., “Appeasement at Munich”). Systematic use of
appropriate historical analogies can clarify the present situation, offer strategic insights,
and inform policy options. On the other hand, incorrectly applying an analogy can
muddy objectives, narrow policy options, and create blind spots for decision-makers.
One can debate when cybersecurity first emerged as an issue, but many consider
the 1988 Morris Internet Worm a common marker. In the 20 years since this self-
replicating program spread across the Internet at remarkable speed, attention has turned
to countering fast-moving, continuously evolving cyber threats and vulnerabilities. In
that time, a single historical analogy has appeared to dominate US Government
thinking: the threat of a cyber Pearl Harbor.
This is a powerful, even seductive possibility. It connotes a bold stroke launched
by an enemy without warning designed to neutralize US military power. This
represents an imminent threat that is ignored only at one’s own peril. The introduction
of new weapons, strategies, doctrines, and tactics suddenly tilt the military balance
toward the offense. Even those who do not directly advocate the Pearl Harbor analogy
D. Sulek and N. Moran / What Analogies Can Tell Us About the Future of Cybersecurity 119
often employ similar imagery. For example, a number of cyber experts have suggested
the potential for a “cyber 9/11”[1] or a “cyber Katrina.[2] While important distinctions
exist between these analogies (e.g., Pearl Harbor centered on a state-based actor, 9/11
on a non-state actor, and Hurricane Katrina on an ‘act of God’), the implications are
clear. Drawing from history’s lessons, experts warn of the potentially catastrophic
dangers facing our cyber networks unless immediate, decisive action is taken.
If the Pearl Harbor analogy proves correct, one can argue the US and other
countries will be better prepared. What if, however, the analogy proves erroneous or
the wrong lessons are drawn? For example, could the focus on a single analogy
ultimately create a self-fulfilling prophecy, something more akin to a modern Guns of
August? This paper will not argue the Pearl Harbor analogy is a bad one. Instead, our
thesis is that while a cyber Pearl Harbor is a possibility, it should not be treated as
inevitable. To test this thesis, this paper will explore a range of historical analogies that
might inform different options and courses of action available to decision-makers.
1. Thinking in Time
An inspiration for this paper is Thinking in Time, which outlined a systematic

framework for policy practitioners to critically analyze key policy challenges and
formulate well-reasoned strategies and options. Through case study analysis, Neustadt
and May point to six problems that often negatively impact the quality of decisions:[3]
• A plunge toward action

• Overdependence on fuzzy analogies
• Inattention to an issue’s own past
• Failure to think about key presumptions
• Stereotyped suppositions about persons or organizations
• Little or no effort to see choices as part of a historical sequence
To address these common shortcomings, the authors conclude that “better

decision-making involves drawing on history to frame sharper questions [about a crisis
or policy challenge] and doing so systematically, routinely.”[3] Specifically, in
response to a crisis or policy challenge, they recommend that decision-makers develop
a detailed issue history. This history will enable them to clarify the overarching policy
objectives and anticipate those conditions that are desired in the future after actions are
taken. At the same time, an issue history provides the basis for determining which
historical analogies might apply—and why. Neustadt and May then outline a process
for developing issue history, summarized below:1
• Determine the Story and Timeline. The centerpiece of an issue history is a

narrative story (what is happening today and why) accompanied by a timeline.
The authors emphasize the timeline should begin at the earliest possible and
relevant date of significance to ensure proper context for analysis.
1
Authors note: we have taken the liberty of condensing and summarizing steps that are spelled out in
detail in Chapters 6-14 of Thinking in Time [3].
120 D. Sulek and N. Moran / What Analogies Can Tell Us About the Future of Cybersecurity
• Identify Change Points. On that timeline, understand where significant

changes altered the trajectory or thinking about the current issue.
• Separate the Known, Unclear, and Presumed. The authors point to the need
early in a policy crisis to determine what is known (facts), what is unclear
(absence of facts or evidence), and what is presumed (assumptions).
• Challenge Presumptions. Decision-makers must carefully review the core
presumptions. Good presumptions are those that clarify and define a situation
and surface concerns. Bad presumptions are value-laden, things that cannot be
challenged save in its own terms by opposed values (e.g., the authors use the
model of “communists are bad; market mechanisms good”).
An issue history becomes the foundation from which decision-makers can

judiciously compare current and past events to determine likenesses and differences.
Embedded in these likenesses and differences are key insights that can shape future
strategies. In concluding their analysis, the authors state, “Sensing that the present was
alive with change, they knew the past would be outmoded by a future that had never
been…but their image of that future could be realistic because [it was] informed by
understanding its sources in the past”[3] In other words, no single historic event will
prove a perfect analogy to the present moment—the underlying conditions will be
different. However, thoughtful selection of historical analogies can offer decision-
makers insights that enrich and inform the choices they must make while also enabling
them to better anticipate the downstream implications of those choices.
2. The Cyber Issue History—In Brief
In April 2007, the Estonian government and many of that country’s key lifeline
infrastructures faced a barrage of coordinated cyber attacks. An unseen adversary
launched sophisticated attacks to cause massive network disruptions—“a flood of
bogus requests for information from computers around the world conspired to cripple
the websites of Estonia banks, media outlets, and ministries for days.”[4] Without
delving into the specific causes, actors, and motives of the Estonian attacks, the entire
event confirmed what many experts warned about cybersecurity. An adversary was
able to employ cyber weapons and strike without warning. Directly attributing the
source of the attack proved fleeting. Critical infrastructures appeared fragile in the face
of withering DDoS attacks. Computer attacks were accompanied by social engineering
and flash mobs to magnify effects. In response to this and other cyber events, the U.S.
launched a comprehensive review of its national cybersecurity strategies.
Today’s cyber issues, however, trace their lineage to the Superpower technology
rivalry that was energized in 1957 with the launch of Sputnik. Following that
psychological shock of this event, the US embarked on an ambitious program to ensure
technological superiority for the foreseeable future. While the Space Program is often
cited as the most significant post-Sputnik achievement, the seeds of the Internet were
planted during this same era. The period beginning in the early 1960s and lasting
through the late 1980s (see Figure 1) was dominated by government, industry, and
academic researchers in their drive to develop internetworking technologies.
Cyber Story, Phase I (1957-1986) Cyber Story, Phase II (1987-present)

• The Story: groundbreaking research to • The Story: emergence of cyber threats
develop new waysto communicate and vulnerabilities coupled with
and collaborate in a resilient way explosive Internet growth/adoption
• The Timeline: • The Timeline:
Sputnik (1957) Computer Security Act (1987)
RAND Study (1962) Morris Internet Worm (1988)
APRANET contact (1968) Cuckoo’s Egg (1989)
First email program (1972) Mosaic (1993)
TCP/IPdevelopment begins (1973) ICANN (1998)
Metcalfe develops Ethernet (1976) Solar Sunrise (1998)
USENET created (1979) Kosovo (1999)
IETFwas established (1986) Code Red (2001)
Estonia (2007)
Figure 1. The Two Phases of the Internet’s Issue History [5] [6]
Events in the late 1980s, however, would introduce vital change points—the
emergence of cyber threats and vulnerabilities. After a raucous debate between the
Executive and Legislative branches over roles and responsibilities for computer
security, President Reagan signed the Computer Security Act in 1987. In 1988, the
Morris Internet Worm is released and quickly self-replicates across the Internet,
causing major disruptions. In response, the Defense Department creates the Computer
Emergency Response Team at Carnegie Mellon University. In 1989, Stanford
University professor Cliff Stoll publishes the Cuckoo’s Egg, which detailed the real-life
penetrations into US systems by a German hacker.
Two mega-trends dominate this second phase. The first is exponential growth in
the number of hosts, users, computing power, and network capacity. For example, in
1984 there were 1,024 hosts worldwide; by February 2008, this number grew to more
than 500 million.[6] Steep, explosive growth curves in these areas were accompanied
by a growing military, economic, and societal dependence on the Internet and computer
networks that permeate nearly every aspect of our lives. The second trend is the
dramatic increase in cyber threats and vulnerabilities. During this period, cyber attacks
grow in terms of velocity (speed of transmission), volume (attack frequency), virulence
(impact, both direct and cascading), and vector (types of actors with the capability to
launch attacks). The 2007 Estonian cyber attacks validated the dangers associated with
the mix of growing dependencies, threats, and vulnerabilities. As one of the most wired
societies in the world, Estonia was particularly vulnerable to this type of attack by a
determined adversary employing hacking tools as the weapon of choice.
Before turning to consider historical analogies that might assist decision-makers
formulate strategies and options for cybersecurity, Figure 2 outlines what we consider
(at a high-level) known, unclear, and presumed about the cybersecurity issue of today.
3. A Framework to Explore Cyber Analogies
The use of analogies is rampant in cybersecurity—and this should come as no surprise.

Neustadt and May note throughout their text that analogies are most often used when
What is Known What is Unclear What is Presumed

• Cyber threats, vulnerabilities, • How grave is the threat? • The United States will retain
and risks continue to grow in • Will next generation Internet a key leadership role in
terms of velocity, volume, technologies and applications governing and influencing the
virulence, and vector be more secure? Internet
• Nation-states and non-state • Is there sufficient political • Nation-states are a more
actors are investing in will (US, global) to address serious threat than non-state
cyberwar capabilities cybersecurity issues? actors
• Decreased resources are • What types of policy • Cyberwar is low risk and high
needed to develop cyberwar approaches (regulation, reward
capabilities market forces, international • Increased public-private
• Internet access and network agreements, others) can cooperation will improve
capacity will continue to change the current security security
grow—with Asia becoming a conditions of the Internet?
more influential actor
• Attribution complicates
response and deterrence
Figure 2. For today’s cybersecurity challenge, what is known, what is unclear, and what is presumed
issues are complex and decisions time constrained. Cybersecurity is an enormously

complex issue with high tech threats and vulnerabilities, a community jargon that
appears to layman as science fiction, attacks that appear with no warning, and a
dizzying array of potential adversaries. Moreover, we live in a fast adapting socio-
technology environment where users routinely change their favorite “killer
applications” on an accelerated cycle, opening new doors of vulnerability.
While Cyber Pearl Harbor is perhaps the most prevalent historical analogy used to
describe the cybersecurity challenge, others (such as Cyber 9/11 and Cyber Katrina) are
being used with increasing frequency. Beyond these, some experts point to the need for
a “Cyber Manhattan Project” or a cyber legal convention modeled after the Law of the
Seas. Still others believe we are in the beginning phases of a “Cyber Cold War”2 and
most recently one cyber expert spoke of the need for a “Cyber Monroe Doctrine.”[7]
Figure 3 depicts a framework to help sort through a plethora of analogies that
might apply to cybersecurity. This model is built along two axes. The vertical axis
divides those analogies motivated by inspiration (hope and possibility) versus those
motivated by desperation (fear and danger). Consider, for example, the contrast
between efforts to deploy the telegraph versus preparations for the Y2K software
vulnerability. The former was motivated by a desire to speed communications across an
unwieldy continent and to facilitate transatlantic communications; the latter driven by a
time-certain fear of a major technological calamity. The horizontal axis divides
analogies where change is systematic (linear, evolutionary) versus those where change
is disruptive (transformative, revolutionary). For example, both the 9/11 attacks and the
outbreak of the First World War were events that significantly altered the course of
history. The former was a disruptive event occurring with little warning and of a scale
not imagined, the latter a product of a system of mobilization and planning that made
war unavoidable once a chain of events commenced (and of a scale not imagined).
2
Panel at the 2009 RSA Security Conference entitled, “Is There A Cyber Cold War?”
Inspiration (Possibilities)
Printing Press Manifest

Destiny
Monroe
Doctrine
Google Telegraph Telephone Nuclear Human

Power Genome
Electrification
Project
Blitzkrieg APRANET Black Sputnik SDI

Markets Space
National Manhattan
Highway Dark Program Project
System Networks
Disruptive Systemic
(Revolution) Law of the (Evolution)
Seas
Pearl Spanish Flu Balkanization Y2K Cold War

Harbor (Influenza
1918)
Katrina
9/11 Outbreak of
WWI (Guns
of August)
Desperation (Dangers)
Figure 3: Framework to Analyze Cyber Analogies.
The remainder of this paper focuses on analyzing four analogies, one from each of
the four quadrants of our framework: the Strategic Defense Initiative, the Cold War, the
National Highway System, and Pearl Harbor. Each will be explored for its likenesses
and differences to today’s cyber issues.
4. The Strategic Defense Initiative (Inspiration, Evolution)
4.1. Overview
During the height of the Cold War, President Reagan proposed the developed of the
Strategic Defense Initiative (SDI). Commonly referred to as the Star Wars program,
SDI was envisioned as a system and capability to destroy Soviet Intercontinental
Ballistic Missiles (ICBMs) while in flight in outer space. This proposed defensive
shield held the potential of negating the carefully constructed logic of Mutually
Assured Destruction (MAD) and conferring the United States a strategic military
advantage over the Soviet Union. The mere threat of SDI forced the Soviet Union to
respond by investing increased resources into its military programs in an effort to
overcome the purported defensive shield constructed by the SDI. While SDI was never
deployed or even proven to be technically feasible, the Soviets were compelled respond
and many analyst believe the economic costs associated with this military buildup in
response to SDI were a contributing factor to the downfall of the USSR.
While SDI provoked an offensive response from the USSR, the threat of cyber
warfare has prodded the US to commit investments toward improving its cyber defense
posture. The Bush Administration’s Comprehensive National Cybersecurity Initiative

(CNCI) is reported to have allocated close to $30B over the life of the program [8]; and
this may, in fact, only represent a fraction of the resources required to protect the US
from a cyber attack of national significance.
Despite the recent cyber attacks against Estonia and Georgia, the threat of large-
scale cyber warfare between states is still theoretical. To counter the potential of this
threat, the US has invested increased amounts of resources (both public and private
expenditures) into cyber defenses designed to protect critical infrastructure—the
purported targets of any cyber attack of national significance. As cyber defenses are not
static and must constantly be monitored, evaluated, and improved in order to counter
determined adversaries, resources must be committed over the long-term. This point is
exacerbated by a fact of life in cyberspace—today, the balance strongly tilts toward the
offense, where the ability to conduct offensive operations is cheaper, easier, and more
effective in comparison to the high costs of mounting credible defenses.
4.2. Similarities and Differences
The similarities between SDI and cyber warfare lie in the responses to perceived threats.
In both cases, the efficacy of the strategies and tactics were unproven. In the case of
SDI, it was not clear the system would ever work—but the USSR could not take the
chance that it might. Similarly, there may be open questions about whether a large-
scale cyber attack might work over a sustained period of time against US military and
infrastructure targets—but the US cannot take the chance that it might. Another
similarity between these analogies is the relative costs of offense to defense. In the
Cold War, the ability to produce nuclear missiles and other delivery systems was
relatively inexpensive and certainly less expensive than trying to develop defensive
systems that would be full-proof. Today in cyberspace, developing offensive
capabilities is inexpensive, especially compared to the enormous costs of developing
cyber defense-in-depth strategies.
The obvious differences between SDI and cyber warfare center on their application.
SDI was inherently defensive in nature, whereas cyber warfare is perceived as
primarily a stealthy, offensive weapon. Further, SDI held the potential to completely
destroy the existing strategic paradigm of Mutually Assured Destruction and
dramatically titling the global balance of power. Cyber warfare is still a new, yet-to-be-
defined strategic paradigm where questions of balance of power are complicated by the
roles and capabilities of governments, private corporations, and a host of non-state
networks of actors (terrorists, organized crime, other dark networks).
4.3. Lessons That Can Be Drawn From This Analogy
SDI never actually worked or was deployed against the Soviet nuclear arsenal. And yet
it has three important lessons for those who seek to develop cybersecurity strategies.
First, it was a program largely motivated by inspiration. Always the eternal optimist,
Reagan sought to find a solution that helped the world escape the horrors of Mutually
Assured Destruction (MAD). Second and related, SDI did not accept the notion that the
offense would always trump defense in the nuclear world. The entire MAD concept
rested on the fact that during any nuclear exchange, both sides would retain sufficient
offensive force to destroy the other. SDI was a bold move to change that paradigm.
Third, in doing so, the US raised the costs for the offense. Today in cyberspace, the
generally held view is the offense trumps defense with cost being the primary
differentiator—it costs billions to erect defense-in-depth in cyberspace, and only
thousands to attack it. However, SDI shows that it may be worth investigating
strategies that seek to significantly increase the costs of the offense rather than trying to
build the perfect defense.
5. The Cold War (Desperation, Evolution)
5.1. Overview
According to McAfee’s 2007 Virtual Criminology Report, we are in the midst of a

“cyber cold war.” Specifically, the report states “attacks have progressed from initial
curiosity probes to well-funded and well-organized operations for political, military,
economic and technical espionage.”[9] The analogy between the modern day cyber era
conflict and the cold war conflict between the Soviet Union and the United States is
primarily anchored in the idea that powerful nation-states are competing for influence
and power without resorting to a direct conventional or nuclear war.
The cyber as a Cold War analogy is ripe with similarities. The most obvious parallel
between the Cyber and Cold War eras is the central role of espionage. The Department
of Homeland Security’s U.S. Computer Emergency Readiness Team received 37,000
reports of attempted breaches on U.S. Government and private sector systems, which
included 12,986 direct assaults on Federal agencies in 2007 [10]. In addition, there
were more than 80,000 attempted attacks on Department of Defense computer network
systems. Countries such as China and Russia have been publicly implicated in many of
these cyber attacks against US military cyber assets. In fact, Major General William
Lord—the Commander of Air Force Cyberspace Command, has publicly stated, "China
has downloaded 10 to 20 terabytes of data from the NIPRNet already." This appears to
parallel efforts during the Cold War, where the Superpowers each invested resources
into the creation and maintenance of rival spy networks. These networks were
primarily designed to gather intelligence in an effort to gain a competitive advantage in
diplomatic, economic, informational, and military confrontations.
Despite these similarities, this analogy is far from a perfect fit. First, the Cyber Era
is multipolar as opposed to the bipolar structure of the Cold War. While the United
States remains an unparalleled superpower, a number of other nation-states are quickly
emerging as potential rivals to the US. In addition, there are a number of non-state
actors (most notably terrorist groups) that threaten to acquire the means to launch cyber
attacks of equal or greater capability that some nation-states. From a military
perspective, this has occurred because the “costs of entry” are low—developing and
maintaining a cyber capability is (in relative terms) remarkably inexpensive.
That stands in stark contrast to the Cold War, where the US and USSR needed to
invest tremendous resources including time, treasure, and knowledge in order to
become nuclear powers and to retain rough technological parity with respect to nuclear
and conventional military forces. According to the Brookings Institute, the US spent
approximately $5.5 trillion dollars on the construction and maintenance of its nuclear
arsenal. The cost of becoming of nuclear power was high in part because of the
tremendous capital investment required in the construction nuclear power plants, the
physical weapons, and acquiring source materials.
In the Cyber era, organizations require only a fraction of these resources to become
a “cyber power.” According to a study conducted by the Naval War College and
Gartner Inc. in 2002, it would require only five years and $200 million to execute a
major cyber attack. [11] As the knowledge and the weapons, in the form of exploit
code, required to conduct a major cyber attack has become increasingly available since
the release of the Naval War College and Gartner study it is likely that such an attack
could be carried out with less resources. The cost of a cyber warfare program is further
reduced because there is very little capital investment required. Unlike nuclear weapons,
cyber weapons are virtual and can be duplicated at very little cost.
The Cold War offers a powerful image, that of a protracted struggle between powers
for political, military, and ideological supremacy. There are obvious similarities—the
cat-and-mouse game of espionage the boils below the geopolitical surface; the proxy
wars that may suddenly break out in cyberspace; and the importance of retaining
technological superiority. There are obvious differences too—the Cold War was an
ideologically-motivated, bipolar struggle between competing nation-states and the fear
of MAD served as a governor on the actions of the two major actors.
However, one important similarity – and lesson to be drawn – is the close
entanglement of economic, political, and security interests in devising a comprehensive
strategy. In the Cold War, the US and USSR brought to bear all instruments of national
power—economic, military, scientific and technological. In particular, the Mr. X
telegram developed by George Kennan at the start of the Cold War outlined a
comprehensive strategy where the US was able to bring all elements of its national
power together toward a common objective, the containment of the USSR. A key
predicate of that telegram was that conflict was inevitable between the two powers, and
the U.S. required a proactive, comprehensive strategy to prepare for the characteristics
of this new conflict. Given the new order being created in cyberspace – where the
Internet touches all aspects of political, military, economic, and sociological life –
perhaps one of the most important lessons from the Cold War is the idea of developing
a Mr. X-like telegram for cyberspace that defines the boundary conditions for future
conflict.
6. The National Highway System (Inspiration, Revolution)
6.1. Overview
In his 1955 State of the Union Address, President Dwight D. Eisenhower declared, “A
modern, efficient highway system is essential to meet the needs of our growing
population, expanding economy, and our national security.” Nearly every aspect of this
ambitious project sought to balance national security, public safety, and commerce as
the country invested billions of federal dollars in road, bridge, and tunnel construction.
The resulting National Highway System represented a remarkable achievement. More
than 50 years after President Eisenhower’s address, our automobile culture has
fundamentally transformed America’s way of life.
A historian looking back 100 years from now might rightly proclaim the National
Highway System and the Internet as two of the world’s greatest cultural achievements,
the Great Infrastructure Wonders of our age. There are numerous similarities. First,
both represented significant step improvements in citizen access and mobility. Second,
both were children of the Cold War. Eisenhower, fresh from his WWII experience in
Europe where transportation and logistics were vital in achieving victory, the National
Highway System was designed in part to ensure the US could quickly mobilize its
forces and deploy them to Europe if the Cold War turned hot. The genesis of the
Internet was, in part, efforts to build a resilient command and control system that could
withstand a Soviet nuclear strike. Third, both infrastructures led to sudden—and
unpredictable—cultural and societal shifts. In less than a decade, the National Highway
System facilitated the growth of suburbs and accelerated the emptying of center cities.
Within a decade, the Internet has created global online communities, perhaps a new
form of “cyburbanization.”
There are two additional, more subtle similarities. First, both infrastructures
emerged from a combination of inspiration and desperation. In selling the idea of the
National Highway System, President Eisenhower often changed his message to suit the
audience. When speaking to war veterans, he emphasized security. When talking to
Chambers of Commerce, he focused on the need to continue post-war economic growth.
When talking to the Rotary Club, he stressed the need to reduce the number of highway
fatalities. While the Internet has its roots in military resiliency, the research community
helped guide and shape its development to promote greater collaboration. Second, both
represented significant shifts in how the United States built infrastructures. Prior to
Eisenhower, the States all developed their own roads and highways with differing
standards and approaches. Eisenhower’s approach “federalized” highways, leading to
unchartered territory in terms of nationwide investments. Similarly, the Internet’s
development is truly unique as an infrastructure. Most US infrastructures began as
regulated monopolies (telephony, power, banking, air travel) and were slowly
deregulated. The Internet has never been regulated—and with its global reach and
impact, has entered equally unchartered territories.
The differences between these two infrastructures are easily identifiable. One was
US-centered; the other is global. The National Highway System was a top-down,
Federally driven program; the Internet is, by its very nature, decentralized and
governance is perhaps best described as ad hoc. Perhaps most germane, the National
Highway System never could be viewed as the avenue of attack against the United
States—it was a force multiplier and enabler. Cyberspace can and has offered some of
the same features to the US as a force multiplier. But cyberspace also introduces
(through technical vulnerabilities in networks) the means by which an adversary may
attack and disrupt critical military and infrastructure operations.
Today, debates in cyberspace are far broader and more encompassing than
cybersecurity. In the United States, we see a need to respond to our vulnerabilities to
growing cyber threats and develop the ability to attribute attacks to better deter, prevent,
and respond to them. At the same time, many of our citizens (and those of other
countries) have expectations of online privacy. Beyond this, since entering office, the
Obama Administration has pushed for greater Internet access, online collaboration,
open government, and transparency. At the international level, we are potentially
entering a new era of Internet governance and influence where other nations share a
common interest in limiting perceived U.S. dominance of the Internet and its
governance structures.
While many of the analogies used today in the U.S. for cybersecurity have at their
core a message of impending danger, President Eisenhower was able to demonstrate
how to strike a balance in his messaging around the National Highway System. His
message was large dose of inspiration—to lower highway fatalities, to create jobs, to
improve the post-war economy—coupled with a tinge of danger, to remain vigilant
should the US have to mobilize to Western Europe in the event of a communist
invasion. Equally important, his approach blended the introduction of revolutionary
concepts (e.g., a stronger Federal role in transportation) with evolutionary steps (e.g.,
the use existing State apparatus’ to facilitate the flow of money). Any significant effort
to address cybersecurity issues will require a similar approach—introducing new
constructs, ideas, and strategies for cyber laws, Internet governance, etc., coupled with
working within the existing confines of the system (at least initially).
The National Highway System analogy also offers a secondary lesson—that these
types of decisions can carry a long tail and produce many unanticipated outcomes. Our
transition en masse to automobiles changed our society (suburbs, summer vacations, an
emphasis on automobile safety), changed our economics (dependence on foreign oil,
rise of trucking), and our environment in ways President Eisenhower could never have
predicted. Ultimately, this makes the case for adopting a balanced approach like that
Eisenhower assumed in the mid-1950s: part inspiration, part desperation; part
revolutionary, part evolutionary.
7. Pearl Harbor (Desperation, Revolution)
7.1. Overview
On December 7, 1941, the Imperial Japanese Fleet launched a surprise attack against
the US fleet anchored at Pearl Harbor. The intent of this attack was to neutralize US
military power in the Pacific as Japan continued to expand its empire.
Many similarities exist between what happened at Pearl Harbor and what many experts
believe theoretically could happen in cyberspace. Among the most obvious is the
introduction of new strategies, tactics, and doctrine. During the First World War,
aircraft were used to perform a variety of military missions, including bombing runs.
As early as the 1920s, Navy’s from across the globe began to recognize the potential of
airpower as a new, offensive form of naval warfare. For example, General Billy
Mitchell theorized that battleships could be sunk via an air bombing campaign. In the
1930s, Japan, the U.S., and Great Britain began to add aircraft carriers to their naval
fleets. Theory was put to the test in November 1940, when the British launched the
“first all-aircraft naval attack in history, flying a small number of aircraft from an
aircraft carrier in the Mediterranean Sea and attacking the Italian fleet at harbor in
Taranto. The effect of the British carrier-launched aircraft on the Italian warships
foreshadowed the end of the ‘big gun’ ship and the rise of naval air power.” [12] The
Japanese studied this raid and built a war plan designed to strike at the heart of US
military power in the Pacific, the US Naval Fleet anchored at Pearl Harbor. One can
argue a similar progression has taken place in cyberspace, from the initial use of
electronic warfare techniques during Operation Desert Shield/Storm through the spike
in hacking attacks during the 1990s and 2000s witnessed in the US Government to the
more coordinated and sophisticated cyber attacks launched against Estonia and Georgia
(the modern day Taranto?).
As noted earlier in this paper, other similarities exist. This includes the notion of
strategic surprise, with an enemy launching a no warning attack with devastating
consequences and the desire to neutralize US military advantages. The general sense
that intelligence and other information exists to point to the attack, information that
may be overlooked or misunderstood if not placed in the correct strategic context.
At the same time, important differences exist in these two situations. First and
foremost, in the case of Pearl Harbor, the enemy and its intentions were well known.
For more than a decade (following the Japanese invasion of Manchuria in 1931),
tensions between the US and Japan spiked. In previous conflicts, Japan had used
strategic surprise to gain military advantage. The US was well aware of Japan’s force
projection capabilities given its large fleet of aircraft carriers. There would be no
confusion about which adversary had determined to strike the United States in the
Pacific. The same cannot be said in cyberspace, where the lack of attribution adds
considerable complexity. Not only is strategic surprise possible in cyberspace, but it is
also possible to veil the source of the attack. Complicating matters, there may be a
number of actors (rival nation-states, rogue states, terrorist groups, and others) with an
interest in not only launching a surprise attack, but potentially even attempting to
stimulate conflict between the victim and a third party. For example, a rogue state
might attempt to launch a large-scale cyber attack against the United States and make it
appear the attacks emanated from another country.
Second, Pearl Harbor required a great deal of lead time and risk for Japanese
planners, moving a giant fleet across half the Pacific Ocean while concealing their
movements. At best, discovery of a large Japanese fleet would have removed plausible
deniability about Japanese intentions. At worst, it could have resulted in military
disaster, as it ultimately did at Midway. In cyberspace, concealment and plausible
deniability are not only possible but relatively easy to achieve and the speed at which
DDoS and other attack techniques can be produced eliminate much of the lead time that
might result in an inadvertent discovery. In other words, strategic surprise in
cyberspace may prove far easier to achieve than in other historic examples, such as
Pearl Harbor or the Israeli attacks commencing the Six-Day War.
A critical lesson to be drawn from the Pearl Harbor analogy points back to a
recommendation by Neustadt and May—understand the timeline and start it at the
earliest possible point. The Pearl Harbor analogy is often used to describe either a
successful surprise attack and/or the failure of a country to anticipate an attack despite
weeks and even months of mounting evidence. Thus, with respect to cybersecurity
today, the analogy is often used to create a sense of imminent danger. But when does
the Pearl Harbor timeline begin? Should one think of the Pearl Harbor analogy
beginning in the fall of 1941 as negotiations between Japan and the US begin to
breakdown? Does the timeline begin with the Japanese invasion of Manchuria, which
signaled larger imperial ambitions? In the 1920s with Billy Mitchell, or over the fields
of Flanders during WWI when aircraft first played critical roles in military operations?
Or does it begin in 1890 when Alfred Thayer Mahan published The Influence of Sea
Power Upon History, 1660-1783, a book that was extraordinarily influential with two
generations of Japanese naval strategists? From the Japanese perspective, does it begin
with the arrival of Admiral Perry and his black ships in 1853?
Choosing the appropriate timeline for Pearl Harbor can greatly change the lessons
cybersecurity strategists might learn from—particularly in identifying relevant historic
parallels and how change points (Taranto, Estonia) altered decision-makers perceptions
and actions.
8. Conclusion: The Power and Perils of Cyber Analogies
Analogies are powerful instruments in the hands of decision-makers. They can: create
and cement an image in the public’s mind about an issue; prove useful in sorting
through the details of a crisis to find key insights; support efforts to develop sound
strategies and well-reasoned policy options; and help leaders anticipate the future,
ripple effects of decisions made today. The paper does not define the “best” historical
analogy for decision-makers to consider in formulating cybersecurity strategies. A
number of factors (e.g., new threats, breakthrough technologies, changing business
conditions) can dramatically alter how the cyber issue may unfold—and, consequently,
which analogies may best apply. The goal of this inquiry was to test our thesis by
creating a framework that puts future cybersecurity events or crises in context—and to
help decision-makers select the most relevant and applicable historical analogies. In
doing this analysis, we’ve reached four conclusions.
First, no single analogy will suffice in considering the complex challenges of
cyberspace. While the cyber Pearl Harbor analogy is rich in imagery, connoting
urgency, it may not be applicable to the full range of cyber scenarios that may confront
the US and the world. Moreover, it holds the potential to produce a dangerous blind
spot—we may wait for the “big one,” a large-scale surprise attack while suffering from
“pinpricks” that, in the end, have a far more debilitating and degrading impact on our
networks (and public confidence).
Second, our examination of historical analogies reveals those that strike a balance
between inspiration and desperation tend to produce the most permanent, lasting results.
The Strategic Defense Initiative, the Manhattan Project, the National Highway System,
and even the creation of the Internet itself have their roots in a mixture of inspiration
(e.g., to destroy incoming ICBMs, to end WWII quickly, to reduce highway fatalities,
and to promote collaboration across a research community) and desperation (e.g., to
counter a growing Soviet nuclear arsenal, to develop the Atomic bomb before Germany
and Japan, to enable rapid mobilization should the Cold War turn hot in Europe, and
the ensure command and control resiliency in the worst case event of a nuclear war).
Third, today many of the analogies used to describe cyber rest at the extremes of
our model. Again, this is not a surprising outcome—these analogies present vivid
images that grab the public’s attention. At the same time, continuously planning only
for the worst-case scenarios can erode public support when these events don’t occur or
their effects are far less than predicted. The Y2K issue offers an excellent illustration of
this. The reality is that early vigilance and a date certain event helped the world prepare
for this software glitch. The perception, however, was that Y2K was anticlimactic and
created more skepticism around gloom-and-doom cyber warnings.
Fourth, one of the most important lessons from Thinking in Time is understanding
the correct, appropriate timeline of an issue: when did this issue start? What are key
change points or trends? The authors argued these timelines should start at the earliest
possible date to fully understand the historical context. For example, one could
examine the analogy of 9/11 in multiple contexts—the year leading up to the attacks; a
timeline starting with the 1993 bombing of the World Trade Centers; or beginning with
the 1982 US Marine barracks bombing in Lebanon. In this case, the same event can
produce different timelines (and resulting historical lessons) that may ultimately
change the focus and core meaning of an analogy. For leaders exploring the future of
cyberspace or dealing with an active cyber incident, starting with the right timeline may
prove critical in making good decisions.
References
[1] National Counterintelligence Executive Joel Brenner in a January 2009 CBS News television interview
(http://www.dni.gov/interviews/20090118_interview.pdf) and by Michael Chertoff, Secretary of
Homeland Security at 2007 RSA Conference
(http://www.theinquirer.net/inquirer/news/1021392/cyber-attacks-significant)
[2] Remarks by Paul Kurtz on February 18, 2009 at the 2009 Black Hat conference in Washington, DC
(http://www.blackhat.com)
[3] Neustadt, Richard E. and May, Ernest R., Thinking in Time: The Uses of History by Decision Makers,
The Free Press, New York: 1986.
[4] Bruno, Greg, The Evolution of Cyber Warfare, The Council on Foreign Relations, February 28, 2008
(http://www.cfr.org/publication/15577)
[5] History of the Internet, http://www.davesite.com/webstation/net-history.shtml
[6] The Hobbes Internet Timeline (http://www.zakon.org/robert/internet/timeline/)
[7] Testimony by Mary Ann Davidson, Chief Security Officer at Oracle, to the United States House of
Representatives Subcommittee on Emerging Threats, Cybersecurity, and Science and Technology,
Source: http://homeland.house.gov/SiteDocuments/20090310143850-78976.pdf
[8] Andy Greenberg, Sketching Obama’s Cyberplans, Forbes, Feb 20, 2009.
http://www.forbes.com/2009/02/20/paul-kurtz-security-technology-security_kurtz.html
[9] Virtual Criminology Report – Cybercrime : The Next Wave, McAfee. Source:
http://www.mcafee.com/us/research/criminology_report/default.html
[10] Jeff Bliss, “Dearth of Technical Experts Leaves US Open to Cyber Attack, Panel Says,” Boston Globe,
March 20, 2009.
[11] Margaret Kane, US Vulnerable to Data Sneak Attack, CNET News, August 13, 2002. Source:
http://news.cnet.com/U.S.-vulnerable-to-data-sneak-attack/2100-1029_3-949605.html
[12] Source: http://wapedia.mobi/en/Battle_of_Taranto
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-132
The Information Sphere Domain

Increasing Understanding and Cooperation
Dr. Patrick D. ALLEN and Dennis P. GILBERT, Jr
Johns Hopkins University, Applied Physics Lab
Booz Allen Hamilton
Abstract. Recent discussions regarding the emerging field of cyber warfare have
focused on the term “cyberspace,” and have included cyberspace as being
considered its own war fighting domain, much like air, land, sea, and space. In this
stage of the Information Age, the international community is grappling with
whether it needs to define this information realm as a domain, similar to the air,
land, sea, and outer space domains that already exist. History shows that there is
always an advantage in a conflict to the side that understands and operates within
a domain better than the opponent. In this paper, the authors propose a definition
of a domain, define what constitutes a domain, posit how new domains are created
over time, and describe the features of what is and is not a domain. These
definitions and features lead to our proposal that the “Information Sphere” should
the preferred international term, and it is this “InfoSphere” that qualifies as a new
domain, with features both similar to and different from the four existing physical
domains.
Keywords. domain, information, cyber, cyberspace
Introduction
Since classical times, two domains of operation dominated military and civilian
operations: land and sea. The advent of powered flight in 1904 initiated the opportunity
for a third domain. Shortly thereafter, actions by opposing elements in this airspace
began during World War I. The Army and Navy each developed its own air capabilities,
and at the end of World War II, the Army Air Corps became the US Air Force—about
50 years after the first powered flight. In a similar manner, the dawn of the “space age”
in 1955 encouraged each of the U.S. military services to invest in their own efforts in
the space domain. By the mid to late 1980’s, with the advent of then US President
Ronald Reagan’s Strategic Defense Initiative (SDI), the US DoD acknowledged outer
space as a fourth war fighting domain.
Based on the preceding observations, the historical trends for recognizing new
domains tend to follow this sequence:
• First, the capability to begin to operate in that domain is developed, such as

the first powered flight or the first space flight.
• Second, the capabilities to operate in that domain become relatively
commonplace, such as air travel or Shuttle launches.
P.D. Allen and D.P. Gilbert, Jr. / The Information Sphere Domain 133
• Third, the capabilities in that sphere to affect capabilities in that domain and in
the other domains become recognized and exploited.
• Fourth, sufficient recognition of the unique and synergistic nature of
capabilities in the domain are recognized and further developed.
• Finally, institutional and financial support for the domain is developed.
1. Definition of a Domain
While considerable dialogue and research has been conducted on the subject, there
does not appear to be an US military definition, NATO definition, or internationally
agreed upon definition for a domain. Joint Publication 1-02, the Department of Defense
Dictionary of Military and Associated terms, does not define a domain. In the absence
of an internationally-accepted definition, we propose a definition of a domain, and
describe features or criteria that distinguish one domain from any other domain in order
to help frame the discussion about whether to define the Information Sphere as its own
domain.
The Webster’s New Collegiate Dictionary has two relevant general definitions of a
Domain:
1. A territory over which rule or control is exercised.

2. A sphere of activity, interest, or function.
Webster further defines a sphere as an area or range over or within which someone
or something acts, exists, or has influence or significance, such as the public sphere.[1]
Note that while the definition of a sphere does include physical environments, it also
includes non-physical environments. Following this train of thought that a sphere or
domain does not have to be a physical domain, it is comprehensible that a sphere can
also apply to area or range in which something acts, exists, or has influence or
significance.
Using these definitions as a guide, we derived the following definition for a
domain for consideration by NATO countries:
The sphere of interest and influence in which activities, functions, and operations
are undertaken to accomplish missions and exercise control over an opponent in
order to achieve desired effects.
By breaking down this definition into its component parts, we can support that
each of the existing four physical domains (air, land, sea, and space) qualifies as a
domain, as defined above. The key components of our proposed definition of a domain
are:
• It is a sphere of interest
• It is a sphere of influence in that activities, functions, and operations can be
undertaken in that sphere to accomplish missions
• It is a sphere that may include the presence of an opponent
• It is a sphere in which control can be exercised over that opponent
134 P.D. Allen and D.P. Gilbert, Jr. / The Information Sphere Domain
Based on the above components, it is clear that the four physical domains of air,
land, sea, and space each qualify as a domain. Each has its own sphere of interest and
sphere of influence. Aircraft fly missions, ships navigate the waterways (both surface
and subsurface), ground forces take and secure terrestrial objectives, and satellites orbit
the earth. In each of these physical domains an opponent can be present and can
interfere with friendly operations. Moreover, the NATO members have military
capabilities in each of these domains, which can be used to control and dominate
2. Features of a Domain
The authors offer for discussion what they consider are the six key feature of a domain:
1. Unique capabilities are required to operate in that domain
2. A domain is not fully encompassed by any other domain
3. A shared presence of friendly and opposing capabilities is possible in the
domain
4. Control can be exerted over the domain
5. A domain provides the opportunity for synergy with other domains
6. A domain provides the opportunity for asymmetric actions across domains
The authors posit that if a domain has these six features, it qualifies as a domain,
and if it does not have all six features, it should not qualify as a domain. This checklist
of features can then be used as criteria to determine whether a new realm, such as the
Information Sphere, qualifies as a domain. The following examples show how the four
physical domains of air, land, sea, and space qualify as a domain according to these six
features:
1. Unique capabilities are required to operate in that Domain. For example,

aircraft are required to operate in the air domain, spacecraft for the outer
space domain, ships for the sea domain, and land systems for the land
domain. Note that each of these capabilities can readily differentiate
themselves from capabilities in other domains.
2. A Domain is not fully encompassed by any other single Domain. For example,
the air domain is not encompassed by the land domain, or vice versa. The
capabilities, missions, and dominance techniques of the capabilities in
each domain remain unique. A tank is not intended to operate in the air
domain, while an airplane is not designed to operate underwater.
3. A shared presence of friendly and opposing capabilities is possible. Any

domain can potentially be entered by opposing forces. This is not to say
that every opponent is present in every domain, but that an opposing
presence must be possible for the sphere of interest and influence to be
considered a domain. A potential shared presence is an essential feature of
a domain since dominance or control over the domain requires the
possibility of an opposing presence or capability.
4. Control can be exerted. The presence of a potential opponent in the sphere of

interest generates the need to influence or dominate such opponents in a
domain. Since a domain is a sphere of influence as well as of interest, then
it must be possible for one side’s influence in a domain to dominate an
opposing side’s influence.
5. Provides opportunities for synergy. The capabilities in a domain must be able

to provide synergistic opportunities with capabilities in other domains.
The classic US Military’s “Air-Land Doctrine” was an excellent example
of how the capabilities of the land and air domains could be mutually
supportive.
6. Provides asymmetric opportunities. Similar to synergistic opportunities are the

opportunities for capabilities in a domain to gain an asymmetric advantage
over opposing forces in other domains. For example, the US Army’s Joint
Fires Doctrine emphasizes the opportunity to use air assets as an
asymmetric threat against opposing land and sea assets, while land or sea
forces can be used to asymmetrically threaten enemy air assets. The
principle of asymmetry must be a possibility for capabilities in a sphere of
interest for it to be defined as a domain.
3. Proposed Definition for the Information Sphere’s Domain
We will next address whether the Information Sphere qualifies as a domain, but first
we have to provide a definition for the Information Sphere. Current DoD doctrine
defines the Information Environment as “the aggregate of individuals, organizations, or
systems that collect, process, or disseminate information; also included is information
itself.”[2] Regrettably, this definition puts the emphasis on the physical attributes of an
information environment. In other publications, the Information Domain has been
described as “the domain where information is created, manipulated, and shared,” or
“where information lives.” These same authors have defined the Cognitive Domain as
the “domain of the mind of the warfighter and the warfighter’s supporting
populace.”[3] With this approach, the content, the connectivity, and the message have
been segregated. We purport that these definitions diverge from the goals of the
Information Operations mission area and the common understanding of Strategic
Communication. Therefore, we first propose a definition of the Information Sphere,
and second, for the Information Sphere’s domain.
The definition we propose for the Information Sphere is:
The space defined by relationships among actors, information, and information

systems.
To further elaborate on this definition, we also define actors, information, and
information systems:
An actor may be a sender, liaison, modifier, transferor, or recipient (either
intended or unintended) of information. Information is the data being passed
among actors via information systems. An information system is any information
perceiving system, information storage system, or communications system,

(including couriers).
Based on these definitions, we propose a new definition for the domain we call the
Information Sphere:
The space of relationships among actors, information, and information systems

that form a sphere of interest and influence in or through which information-
related activities, functions, and operations are undertaken to accomplish missions
and exercise control over an opponent in order to achieve desired effects.
Note that the information by itself is not the domain, nor is the domain simply the
information systems in which the information rides and resides. It is the space defined
by the relationships among actors, information, and information systems that define the
Information Sphere and allow it to qualify as a domain. We include the word “sphere”
in the Information sphere to distinguish the fact that the domain we are proposing
consists of more than just the information component, and calling it the “information
domain” would encourage that misunderstanding. The other accepted domains do not
use the term “the air sphere” or the “sea sphere,” but we use “Information Sphere” to
make the distinction from just the information component completely clear.
This definition is different from the previously referenced definitions for the
cognitive, information, and cyber domains because the proposed definition of the
Information Sphere explicitly includes the relationship among these three components.
It is the relationships among these three components that define the meaning, context,
and value of the Information Sphere, not the three components taken in isolation.
The US Military’s Quadrennial Roles and Missions Review Report defines
cyberspace as “a global domain within the information environment consisting of the
interdependent network of information technology, infrastructures, including the
Internet, telecommunications networks, computer systems, and embedded processors
and controllers.”[4] We believe that this is a good definition of cyberspace, but believe
that cyberspace is still a subset of the larger Information Sphere domain. Just as naval
surface actions and submarine actions are two components of the Sea domain,
cyberspace, cognitive, and information are components of the more encompassing
Information Sphere.
4. The Evolution of the Information Sphere as a New Domain
As mentioned above, there are five steps that capabilities in an environment tend to
follow en route to becoming a new domain:
1) The capability to begin to operate in that domain is developed. From the advent
of the PC and the birth of the public version of the Internet, communication and
information capabilities have exploded. Combined with global transportations, these
capabilities provide a global economy and social interactions to a degree previously
unheard of.
2) The capabilities to operate in that domain become relatively commonplace.

Fourth Generation cell phones, PCs, and Internet access are now commonly found
almost everywhere in the world. Nations that have yet to develop their communications
infrastructures are jumping straight to fourth generation access that does not require the
construction of expensive information infrastructures. Almost anyone in the world can
achieve global communications via the Internet and cell phones.
3) The capabilities in that sphere to affect capabilities in that domain and in the
other domains become recognized and exploited. Information has always been
important to military and civilian operations. The Information Age has made the
Information Sphere not only widespread but also shared. Opponents can reach our
internet-connected networks without leaving their own country. There are few places in
the world where the news media do not reach. Incidents in the remotest parts of the
world often carry global implications beyond any time in previous history. As a result,
conflict in the Information Sphere is becoming more prevalent and more important than
even direct military action in many cases.
4) Sufficient recognition of the unique and synergistic nature of capabilities in the

domain are recognized and further developed. As both the capabilities and threats in
this sphere continue to grow, more and more resources are being allocated to the
exploitation and securing of Information Sphere capabilities.
5) Sufficient institutional and financial support for the domain is developed. The
US’s efforts to create a new Sub-Unified Command for Cyber is one example of efforts
toward developing the necessary institutional and financial support for operating and
succeeding in the Information Sphere. The future may be an US Interagency
organization, or even an international megacommunity, that represents the Information
instrument of national power, along with the Diplomatic, Military, and Economic
instruments. Note that the bringing together of Information Sphere capabilities from the
other instruments of national power (including military) is the logical progression we
would expect to see as the Information Sphere domain becomes institutionalized and
supported financially.
5. What is Unique About the Information Sphere’s Domain?
Now that we have described why the Information Sphere qualifies as a domain, this
section will describe why the Information Sphere is also unique compared to the four
physical domains. At the same time, we will describe why we believe that what makes
the Information Sphere unique is yet one more reason why the Information Sphere
should be treated as a new domain.
Since the definition of the Information Sphere includes actors, information, and
information systems, it is evident that each of these three components must reside in a
physical medium at any point in time. For example, an information server is located
either on the ground, underground, in the air, in outer space, on the sea, or (potentially)
under the sea.
In a similar manner, the information itself is either being stored on an information
system, or is in some information conduit (including a portion of the electromagnetic
spectrum) at any point in time. Finally, the human actors must be located in one of the
four physical domains. Figure 1 gives an example of how one might consider the
information sphere’s domain with respect to the four physical domains.
Note that this figure shows that the Information Sphere is separate from each of the
four physical domains, but is also accessible by, and provides access to, all four.
Information may enter or exit via a physical medium, but that may or may not be
relevant. For example, if an intruder is seeking an entry point into a network, the
physical location of the entry point may matter to the intruder. However, once the
intruder is in the network, the physical location of the entry point and any informational
areas of interest are of less importance due to the degree of access provided by the
network. What is important in this case is the relationship between security elements of
the network (including people), the targeted information, and the intruder, rather than
the physical location of the assets.
Space
Air
Information
Sphere
Land
Sea
Figure 1. Information Sphere is a Unique Type of Domain[5]
Note that the concept of entry and exit points from one domain to another is
prevalent in all domains. Aircraft and spacecraft land on the ground or at sea. Ships
dock at land-based ports. The same is true for the Information Sphere. There will
always be entry and exit points from the Information Sphere to and from the other
domains, as the purpose of most activity in the Information Sphere is to affect
something in the physical world. However, there are also actions and desired end states
associated with operations within the Information sphere that are unique to the
Information Sphere, irrelevant to and unaffected by the physical space in which the
actors, information, or information systems actually reside.
Each domain has actions that are dependent and independent of each of the other
domains. Similarly, the Information Sphere is not completely encompassed by any
physical domain. For example, a distributed database that has elements either residing
or in transit on land, in the air, on the sea and/or in outer space is not contained or fully
encompassed by any of the four media in which it is located or passes through.
Moreover, even the union of air, land, sea, and space physical environments does
not fully encompass the Information Sphere. The interactions we described for the
Information Sphere often occur in a space of relationships where the physical location
of the actual components is irrelevant once access has been achieved. For example, the
ability for two actors to interact in some way does not depend on the medium or media
within which the information exchange occurs. What matters are the interaction and the
relationship between the actors, information, and information systems? Moreover,
shared presence within all four physical domains does not equate to dominance in the
Information Sphere, either in the control of information access, information systems, or
in the beliefs and perceptions of groups of interest within those four domains.
It is these relationships between actors, information, and information systems that
define the interest and influence mechanisms in the Information Sphere. Since these
relationships can be satisfied by a wide range of paths into, out of, and through various
physical media, the value, benefit, and vulnerability of elements within the Information
Sphere are relatively independent from the four physical domains.
Another important distinction is that the desired effects of an information activity
eventually reside in one or more of the four physical domains. For example, the
information activity may be to bring down an enemy air defense system, which opens
the way for the air operations, which shapes the upcoming ground or sea battles.
However, there may be a significant delay between the initial information activity and
any effect in one or more physical domains. For example, the placement of a back door
on a target server does not have an immediate effect other than the opportunity for
access at a later date. Until that access is exploited, there is no physical manifestation
of a desired effect.
As another example, competition between opposing thoughts or beliefs frequently
has a delayed reaction. The concept of freedom, for example, is often dormant until the
opportunity to be free, or to achieve increased freedoms, becomes available. In the
conflict among beliefs, a thought that is planted may blossom many years later after
additional thoughts and physical events have occurred.
Therefore, the fact that the actors, information systems, and information that
comprise the Information Sphere must reside at any instant in one of the four physical
domains is either secondary or irrelevant to the functioning of the abstract relationships
within the Information Sphere. Information easily transcends the barriers between the
physical domains. The Information Sphere is a space where the understanding of
relationships in that space can lead to dominance over opponents in that space.
6. The Information Sphere’s Qualifications as a Domain
We argue that the Information Sphere qualifies as a domain according to our preceding
definition for the following reasons:
• The space of relationships among actors, information, and information

systems forms a sphere of interest
• It is a sphere of influence in that activities, functions, and operations can be
undertaken in that sphere to accomplish missions
• An opponent to friendly operations may function in that sphere
• Control can be exercised over that opponent in or through that sphere
Besides meeting the four preceding criteria described above, the Information
Sphere also meets each of the six features required of a domain.
1. Unique capabilities are required to operate in that Domain. Information

capabilities are required to operate in the information realm. The Information Sphere
requires unique equipment and personnel skills to function effectively, accomplish
missions, and dominate any enemy presence. Information capabilities operating in the
Information Sphere are both unique and differentiable from the capabilities designed to
operate in other domains. For example, a computer system (and associated
software/code) optimized for hacking into enemy computer networks is a unique asset
different from air, land, sea, and space platforms. Hacking skills are unique from the
more traditional set of pilot, sailor, soldier, and astronaut.
As information capabilities become more specialized, the uniqueness and
differentiability of these capabilities will continue to grow. For example, the
Information Sphere now has a set of unique equipment (materiel) and personnel skills
required to effectively operate in, defend, and attempt to dominate the domain. With
these new capabilities comes a range of unique support structures, such as doctrine,
organization, training, leadership development, facilities, and policy.
2. A Domain is not fully encompassed by any other single Domain. The

Information Sphere is not fully encompassed by any combination of land, sea, air, or
space domains. The Information Sphere has capabilities and functions that are
meaningful only in this information environment.
3. A shared presence of friendly and opposing capabilities is possible. Until

recently, the Information Sphere rarely allowed for a shared presence. A shared
presence was not feasible primarily due to physical and geographical separation and the
inherent time delays. With the birth of the Information Age, however, the Information
Sphere is frequently shared. Examples of this sharing include the range of information
media, including the Internet, local and wide area networks, television, radio, print
media, video and audio recordings, and other capabilities. Due to the explosion of
information and information capabilities, the Information Sphere allows for a shared
presence more than ever before. As a result, dominance and control in this domain have
become much more important than in the past.
4. Control can be exerted. For the Information Sphere, control can refer to the
control of the information systems in a region of the information sphere, control to the
access of the information in that information sphere, or even the dominance of one
belief over another in a region of the information sphere. As an example, air-to-air
radars on fighter aircraft may try to jam or spoof the radars of opposing forces in the air
domain, thereby attempting to control the information sphere. The recent spate of
alleged nation-state sponsored hacks into sensitive but unclassified US military and
contractor information systems is an example of the type of temporary but useful
control our opponents can undertake in the Information Sphere.[6] Influence over
population groups is a constant competition in the Idea Battlespace among ideas vying
for dominance over other ideas.[7]
5. Provides opportunities for synergy. The Information Sphere provides synergistic

support to all the other domains, and vice versa. The ability to gather information
directly from an enemy information source can assist air, land, sea, and space
operations. Conversely, the ability to take out an enemy information system from the
air can force the enemy to use an information system already compromised by our
side.[8]
6. Provides asymmetric opportunities. Information capabilities can provide an

asymmetric threat against enemy capabilities in other domains. In his book, The Next
World War, James Adams[9] describes a case where a computer virus was entered into
a printer that was supposed to be delivered to an enemy site in order to neutralize an
enemy air defense system. In a similar manner, physical assets can be used to disrupt
and destroy opposing information systems.
7. Benefits of Treating the Information Sphere as a Domain
First, there is always an advantage in a conflict to the side that understands and
operates within a domain better than the opponent. This is true on land, air, sea, and
space, and can also be true in the Information Sphere. Obtaining dominance in the
Information Sphere will likely lead to continued dominance in the four physical
domains via asymmetric effects. By defining the Information Sphere as a domain, a
body of knowledge or military science of operating in the Information Sphere will be
more thoroughly developed to improve understanding and consensus on the subject.
Second, representing the relationships of information among actors and

information systems in a manner useful to planners and decision makers will help
improve the effectiveness and efficiency of operations in and through the Information
Sphere. For example, the ability to readily visualize relationships in a common format
will facilitate a unity of effort and common understanding of objectives and constraints.
Defining the Information Sphere as a domain should lead to an investigation and
experimentation on a number of methods to represent these relationships, and the best-
of-breed methods should emerge to enhance our capabilities in this domain.
Third, focusing and preparing enhanced capabilities in the Information Sphere

will enable superior influence and control in this domain. The side with better
personnel, equipment, doctrine, organizations, and leadership will have a significant
advantage over the opposition. If the military departments of NATO countries choose
to define the Information Sphere as a formal war fighting domain, then the resourcing
to more effectively and efficiently function in that domain should follow.
Fourth, defining the Information Sphere as a domain allows for increased emphasis
on planning and employing all instruments of national power—diplomatic,
informational, military, and economic—in a common, coordinated endeavor. Since
information is a common element in the use of all instruments of national power, the
ability to function effectively in this domain will encourage the coordination and
synchronization of effects among all these instruments.
Fifth, defining the Information Sphere as a domain will help increase the emphasis
on improved information assurance and cyber security, which can and should lead to
improved economic and national security. Defining the Information Sphere as a
domain will help define the common areas of interest between these sectors, and
eventually lead to common, or at least coordinafted, resourcing in the areas of

information security.
Finally, defining the Information Sphere as a domain can help focus international
efforts on the important conflicts already ongoing in this domain. In addition to the
skirmishes in cyberspace, the battle for the hearts and minds of many groups of actors
worldwide has been raging since the birth of philosophies and political systems. In a
battle of the minds, the physical location of the people believing in something is less
important than the dominance of that belief over other competing beliefs. Defining the
Information Sphere as a domain will help highlight the need for renewed effort and
capabilities in this cognitive realm.
8. Summary
This paper presented the definition and features of a domain, a definition for the
Information Sphere, and why the Information Sphere qualifies as a domain along with
the four physical domains of air, land, sea, and space. The paper also presented why the
Information Sphere has some distinct differences from the four physical domains, and
the benefits of treating the Information Sphere (which includes cyberspace) as its own
domain. Lastly, the paper describes why the Information Sphere is a comprehensive
domain that encompasses the areas of cyberspace, cognition, personnel, and
information itself, which is why we include the term “sphere” in the definition.
Referring to this new area as the “Information Domain” would imply that the focus of
this domain is focused just on the information component, which does not adequately
capture the full scope of the new proposed domain: the “Information Sphere Domain.”
References
[1] http://www.merriam-webster.com/dictionary/sphere; accessed 14 January 2009.

[2] Joint Publication 1-02, Department of Defense Dictionary of Military and Associated Terms
(Washington D.C.: U.S. Government Printing Office, 12 April 2001), page 203.
[3] Alberts, David S., Gartska, John J., Stein, Frederick P., Net-Centric Warfare-Developing and
Leveraging Information Superiority (DoD C4ISR Cooperative Research Program, 2nd Edition
(Revised) August 1999.)
[4] Gates, Robert M., Quadrennial Roles and Missions Review Report, Department of Defense, January
2009, http://www.defenselink.mil/news/Jan2009/QRMFinalReport_v26Jan.pdf, accessed 1 February
2009.
[5] Allen, Patrick D., Information Operations Planning (Artech House, New York, 2007), p. 298
[6] Wortzel, Larry M.,Chairman, et al., 2008 Report to Congress (U.S.-China Economic and Security
Review Commission, Washington, 27 October 2008), http://www.uscc.gov, accessed 25 November
2008.
[7] Allen, op cit., p. 114
[8] Koch, Andrew, Information warfare tools rolled out in Iraq, (Janes’ Defense Weekly, Washington, 6
August 2003)
[9] Adams, James, The Next War, Simon and Schuster, New York, 1998.
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-143
Sun Tzu was a Hacker: An Examination of

the Tactics and Operations from a Real
World Cyber Attack
Billy K. RIOSa
a
GreyLogic., LLC
Abstract. This text will cover the operational and tactical techniques used in a
“real world” cyber-attack and includes an analysis of the planning, command,
control, execution, and outcome of these cyber-attacks. The text focuses on the
cyber-attacks against the nation state of Georgia in 2008, as the author was in a
unique position to observe the communications, execution, and responses from
both attacking and defending entities. The various aspects of the attacks will be
described and linked back to traditional concepts of Maneuver Warfare as
described in Marine Corps Doctrinal Publication 1 (MCDP-1).
Keywords. cyber warfare, cyber attack, offensive cyber capability, defensive

cyber capability, Georgia, maneuver warfare
Introduction
“Doctrine must evolve based on growing experience, advancements in theory, and the
changing face of war itself.”
General C.C Krulak, MCDP-1
The computer system sitting on your desk brings a new dimension to modern warfare.
Just as the machinegun and airpower changed the face of warfare, so does offensive
cyber capabilities. The ubiquity of computer systems in our military, government, and
civilian infrastructure have solidified the importance of offensive cyber capabilities to
the point where packets will be the “bullets” that will be fired in future conflicts.
Software loaded on computer systems will be considered the “terrain” on which cyber
warfare is waged. The use of packets as weapons presents a novel approach to warfare
and will inevitably cause changes in military doctrine within our lifetime; however the
employment of cyber capabilities continues to abide by many of the traditional
concepts and principles of warfare. Associating the concepts of cyber war in the terms
of conventional warfare and known doctrine can make analysis less daunting and
provides a new perspective when measuring the impact and deciding how best to
employ cyber capabilities. In this text, the author will examine the execution of a
cyber-attack and correlate the principles of the attack to traditional concepts studied in
maneuver warfare. The specific scenarios examined in this study involve the cyber-
attacks conducted against the country of Georgia in 2008. Although the author uses a
144 B.K. Rios / An Examination of the Tactics and Operations from a Real World Cyber Attack
specific cyber-attack to illustrate his points, the author hopes that the concepts
presented in this text can be universally applied to any cyber-attack.
1. Maneuver Warfare
The principles of maneuver warfare will be referred to extensively in this text. While a
complete description of all the concepts associated with maneuver warfare are beyond
the scope of this text, the author begins by providing a fundamental description of
maneuver warfare and its foundations. According to the Marine Corps Doctrinal
Publication 1 (MCDP-1), maneuver warfare is described as a:
“war fighting philosophy that seeks to shatter the enemy’s cohesion through a variety
of rapid, focused, and unexpected actions which create a turbulent and rapidly
deteriorating situation with which the enemy cannot cope”
The focus on disrupting enemy cohesiveness makes maneuver warfare unique

from other military doctrine. In maneuver warfare, entire enemy units and strongholds
are bypassed in order to reach a “decisive opportunity” to exploit a “critical
vulnerability” in the enemy’s position [1]. Exploitation of critical vulnerabilities
provides a pathway to attacking the enemy’s “center of gravity” [1]. It is important to
understand that despite its name, maneuver warfare is not limited to the maneuvering
of units or spatial operations. Temporal actions such as psychological and
technological disruption are also key elements of maneuver warfare. In maneuver
warfare, well timed combat power brought to bear on strategic points on the battlefield,
preemptive strikes geared towards the elimination of the enemy’s decision making
ability, surgical strikes on communication systems, elimination of the enemy’s
logistical chains, and suppression of enemy combat power are all more highly valued
over high body counts or gained geography [1].
While the physical destruction of enemy forces and equipment is not the primary
focus of maneuver warfare, physical destruction and firepower do play a central role at
decisive points in battle, especially when destruction of enemy forces results in the
degradation of the enemy’s overall cohesion. Advanced weapon systems and technical
superiority such as superior weaponry, stealth technology, and highly trained special
operations forces (SOF) can increase the aggressor’s opportunities to deliver decisive
firepower on the right targets at the right moments, disrupting the enemy’s normal
operating rhythm and decision making ability. For example, in the lead up to Operation
Desert Storm in 1991, SOF and air power disabled and destroyed a significant portion
of the Iraqi command and control systems, disrupting Iraqi command and control at the
highest levels [2]. While the success of the air power and SOF units would not have
“won the war” in isolation, they disrupted the enemy’s cohesion and decision making,
allowing for more effective follow on operations which would ultimately win the war.
It is this “disruptive capability” that is the quality that makes offensive cyber
capabilities so attractive. The ability to disrupt the enemy’s tempo, rhythm, and
decision making from afar, in a lighting fast manner, while exposing very little, is
extremely appealing to many commanders.
Although offensive cyber capabilities offer a novel approach to disrupting the
enemy’s normal rhythm and decision making, the prudent commander understands that
much like air power, naval power, intelligence, and other individual military
B.K. Rios / An Examination of the Tactics and Operations from a Real World Cyber Attack 145
capabilities, offensive cyber capabilities cannot “win a war” by itself [3]. Instead, these
offensive cyber capabilities must be used as a component in the overall combined arms
effort focused on disrupting the enemy’s cohesion and exploiting critical vulnerabilities.
Once the enemy’s battle rhythm and decision making is disrupted, the disruption must
be exploited via follow on actions. Disruption creates opportunities that should be
ruthlessly exploited. This exploitation often leads to additional opportunities, which
eventually leads to a decisive opportunity to launch a decisive attack against the enemy
[1].
The author will present specific scenarios where offensive cyber-attacks were used
in a manner that was consistent to the principles of maneuver warfare. The author
chooses to focus on the 2008 cyber-attacks launched against the nation state of Georgia.
These attacks are chosen due to the resources and vantage points the author held whilst
the conflict progressed [4]. These resources and vantage points gave the author an
insight into both sides of the conflict; however most of the examples and scenarios will
focus on the aggressor and offensive actions. Before diving into the specific attacks
that occurred against the Georgia infrastructure, it is important to define several terms
which are used regularly in describing maneuver warfare. The author chose to focus on
the following terms throughout the course of the text.
• Decentralized Command and Commanders Intent

• Combined Arms
• Initiative
• Centers of Gravity and Critical Vulnerabilities
1.1. Commanders Intent
The commander’s intent provides the means for subordinates to exercise judgment and
initiative. MCDP-1 states that each mission has two parts: (1) the task to be
accomplished and (2) the reason or intent behind it. Commander’s intent provides the
reasoning and intent behind the assigned tasks and missions. Commanders intent is
crucial for as the situation changes on the battlefield, the specific tasks assigned to the
subordinate may become obsolete, but the intent is lives beyond the assigned tasks and
continues to guide the subordinate’s actions [1]. If the subordinate understands the
commander’s intent, they will be able to execute actions without the presence of direct
orders and those actions will be in line with the commander’s desires (which should
ultimately advance strategic objectives).
In addition to the promotion of initiative, effective use of commander’s intent
allows for decentralization of command, pushing decision making to the lowest level. It
is at these levels where forces are able to react and exploit opportunities in the most
effective and efficient manner. Decentralized command and asynchronous execution is
essential in the success of conventional operations in today’s “small wars” [5] as well
information based campaigns.
1.1.1. Target Lists and Commanders Intent

In August of 2008, the Grey Goose project commenced. Grey Goose was a pure open
source intelligence initiative aimed at gathering and analyzing intelligence related to
the Georgia cyber-attacks. During Phase I of the Grey Goose project, a number of
Russian hacker forums were mined for data detailing over 29,000 separate forum
events with correlation of those events to status of Georgia cyber infrastructure [4].
One of the first items discovered on the various Russian hacker forums were target lists,
providing the domain names of various Georgian servers to be attacked. A portion of
the target list is shown in figure 1.
Figure 1. Target list from Russian hacker forum
The target lists discovered on the various forums were simple and provided no
specific direction or instruction as to how the various sites were to be attacked. The
targets lists were simply lists of servers that should be targeted and attacked by forum
members. Forum members were not assigned specific tasks, the specific techniques to
be used were not defined, and the definition of a “successful attack” was broad and
conceptual. Instead of publishing specific actions, the publisher of the target list
allowed the forum members to decide the best course of action to carry out the attacks.
Soon after the target list was posted, the forum was filled with chatter related to the
most effective means of attacking the various servers. The communication was not
directed to “higher” (to the forum administrator) asking for guidance, but instead
focused on “lateral communication” (to other forum members), updating each forum
member on newly discovered vulnerabilities and weaknesses [6]. As the lateral
communications increases, each forum member cherry picks the data most appropriate
to their interests and skillsets. This prejudicial filter helps maximize the impact of each
individual forum member by allowing them to focus on applying their specific skillsets
to attacking the servers on the target list, quickly identifying those servers that are most
vulnerable to those specific skills possessed by the individual. An individual

contribution to the forum is shown in figure 2.
Figure 2. Various Forum Members Discussing Vulnerabilities
Despite its lack of detailed instruction, the target list established the framework for
the most effective use of forum member skillsets. The target list and broad concepts of
“success” are prime examples of an effective use of commander’s intent [7]. As
opposed to identifying individual forum members, attempting to determine their skillset,
and assigning the forum members specific tasks/servers/exploits, the forum
administrator establishes the overall intent of the attacks and essentially publishes the
intent. The forum administrator publishes the target list (task to be accomplished) and
broad guidance (the reasoning/intent behind the task). The forum administrator
provides little/no guidance as to how the tasks are to be accomplished. Instead, the
forum administrator relies on the forum members to develop their own methods to
accomplish the tasks within the overall intent, exploiting weaknesses as they become
evident, and relaying updates through lateral communications. As the situation changes
(new vulnerabilities discovered, new defenses encountered, new tools released…etc.)
the forum members proceed within the original intent and do not wait for further
instruction. This allows maximum flexibility and effectiveness in attacking, a
flexibility that simply cannot be matched in a highly centralized, top down command
and control structure [1].
1.2. Combined Arms
Combined arms are utilized to maximize combat power. The term “combined arms”
refers to making use of all the available resources to the best possible advantage.
Combined arms are typically achieved through the complementary use of different
weapon systems [1]. The weaknesses of one weapon system are supplemented by the
strengths of a different weapon system. The classic example of combined arms
automatic direct fire weapons (machine guns) and indirect fire weapons (grenade
launchers). If the enemy infantry becomes pinned down by the automatic fire, they
become vulnerable to grenade attacks. If the enemy maneuvers to avoid the grenade
attack, they expose themselves to the automatic weapons fire. The ultimate goal of
combined arms it to utilize a full integration of various arms to achieve a situation so
that when the enemy counteracts one arm, they are making themselves more vulnerable
to another [1].
1.2.1. SQL Injection, DDOS Tools, and Combined Arms
In order to extract the maximum effect from offensive cyber strikes, the strikes must be
used as part of a combined arms effort. This combined arms effort can involve the
leveraging the exploitation a single cyber related vulnerability to accomplish successful
exploitation of another cyber related vulnerability. The combined arms effort could also
involve the exploitation of cyber related vulnerabilities in conjunction with the use of
kinetic weapons or conventional forces. During the investigation of data made available
to the Grey Goose project, exploitation of several SQL injection vulnerabilities against
various Georgia applications were discovered [4]. SQL injection attacks in the logs of a
Georgia server are shown in figure 3.
Figure 3. Examination of log files show evidence of SQL Injection attacks
These SQL injection vulnerabilities were initiated from Russian IP addresses and
log data provides many of the exact SQL injection queries that were used in the attacks.
These targeted SQL injection attacks began in July of 2008, months before the high
profile attacks against Georgia in August of 2008. The SQL injection attacks started
with simple fingerprinting of the backend database servers being used by the
vulnerable applications. An examination of the log files shows that once the
fingerprinting of the backend database was complete, the Russian hackers extracted the
usernames and passwords associated with the vulnerable applications. The usernames
and passwords are valuable because they provide the foundation for attacks against
other systems. For example, once the usernames and passwords are extracted, the
hacker can test those username and password combinations against other, better
protected information systems (password reuse) [8]. If any of the users of the
compromised application have reused their password on other systems, the hacker can
now masquerade as a legitimate user on that other system. Password reuse can also lead
to the compromise of personal and business email accounts, providing a stream of
intelligence that can be used in conjunction with other attacks (both cyber and
conventional). If a hacker has gained access to the business and personal email systems
of those employees, the hacker will be in a prime position to collect intelligence on
those individuals, feeding the captured data into traditional intelligence analysis and
fusion.
Figure 4. Forum Members Make DDoS Tools Available to Novice Hackers
Once the attacks against Georgia became public and conventional action was
imminent, many of these SQL injection vulnerabilities were posted on various hacker
forums, allowing others to take advantage of the SQL injection vulnerabilities. In
addition to the SQL injection attack vectors, tools were developed and posted to enable
the flooding of Georgia servers, creating a distributed denial of service attack against
various parts of the Georgia infrastructure. The automated tools gave even novice
hackers the ability to disrupt communications while also spreading the attacking
surface. An example of the tools being distributed is shown in figure 4. Both highly
skilled, targeted attacks coupled with unskilled, broad DDoS attacks were used against
the Georgia infrastructure, forcing the defender to address both issues simultaneously.
Once attention and resources were dedicated to defending the information systems, the
Russian military began the conventional campaign [9].
The exploitation of a single application level vulnerability leads to further
compromise and exploitation, long after the initial vulnerability is fixed. Pilfered
information leads to more pilfered information, which in turn leads to even more
information theft and swells into enormous amounts of sensitive data being stolen. This
data is fed into traditional intelligence analysis efforts, helping to paint the picture of
the lives of government employees involved with various projects on separate,
seemingly unrelated servers [10]. As the situation changes, some vulnerabilities can be
repurposed to fit new situations and scenarios. Various attacks (SQL injection,
Command Injection, DDoS…etc.) are launched simultaneously, forcing the defender to
addresses multiple scenarios and skillsets simultaneously. Eventually, the Russian
military initiated contact though conventional means, adding yet another dimension
onto the defenders dilemma [9]. This chain of events began with the targeted
exploitation of a single vulnerability in a single application and grew into multiple
attacks launched simultaneously, along with the beginning of a conventional campaign.
1.3. Initiative
MCDP-1 describes two states that all actions in war are based upon; “initiative” and
“response”. MCDP-1 describes initiative as the ability to “dictate the terms of the
conflict and force the enemy to meet us on our terms” and response as “resistance to
initiative” [1]. Taking the initiative is considered the more preferable of the two states
as, “it is through the initiative that we seek to impose our will on the enemy” [1]. In
traditional military operations, initiative is established by forcing the enemy to assume
a reactionary stance against active actions. These actions typically gain initiative
though the effective use of surprise, tempo, concentration, and audacity. Offensive
cyber capabilities offer tremendous opportunities to gain the initiative [11]. The very
nature of cyber-attacks brings about the elements that are commonly associated with
initiative: surprise, tempo, concentration, and audacity. Initial contact can be initiated
with little risk as attacks can be launched from a variety of locations, including non-
state sponsored educational and commercial networks. The nature of today’s
networking makes uncovering undisputable links to State sponsorship an extremely
difficult task [12]. These attacks have the ability to disrupt conventional systems and
decisions making from long ranges, helping shape the battlefield well before any
rounds are fired.
An example of how cyber-attacks can be used to help establish the initiative for
conventional forces is presented in the attacks against Georgia in 2008. In July of 2008,
before a single shot was fired by conventional forces, Russian based hackers had
already penetrated Georgia government applications with SQL injection and other
application level attacks [13]. During the pre-emptive cyber strikes, several high profile
systems (such as the website of the President of Georgia) were compromised. The pre-
emptive attacks undoubtedly captured the attention of the Georgia government, forcing
parts of the Georgian government to utilize a “decision making cycle” in order to
determine the appropriate response [14]. By initiating contact and forcing the Georgian
government to enter the decision making cycle, the Russian hackers gain the initiative,
forcing a reactionary stance by the Georgian government. Much like conventional
attacks, a single, un-sustained attack is insufficient in maintaining the initiative, so the
Russian hackers followed with sustained attacks against a wide range of government
systems. Eventually, these cyber-attacks were followed by conventional ground and air
attacks. Each phase of the attack is meant to keep the enemy off balance and in a
reactionary state. Agility, tempo, and surprise continuously disrupt the defenders
decision making, allowing the attacker to dictate the terms of engagement. As the
defenders observe, orient, decide, and attempt to act upon targeted attacks, the attackers
launch broad denial of service attacks against the entire infrastructure. As the defenders
rush to observe, orient, decide and act to defend the wide scale cyber-attacks, the
attacker changes the terms of engagement and initiates the conventional ground and air
campaign. These offensive cyber-attacks were not the ultimate end state; instead they
were used to augment the achievement of initiative in conventional warfare in support
of the true main effort.
1.4. Centers of Gravity (CoG) and Critical Vulnerabilities
Building a combat capability is not sufficient to win a war; to win a war, the built up
combat capability must be directed towards a decisive objective. Although several
interpretations for Centers of Gravity (CoG) exist, MCDP-1 considers CoG as the
“sources of strength for the enemy”. These sources of strength need not be physical and
can encompass “intangible characteristics such as resolve or morale”. MCDP-1 states
that centers of gravity are to be attacked (although not directly, if well-defended).
While CoG focuses on the enemy’s strengths, critical vulnerabilities focus on the
enemy’s weaknesses. While the enemy is likely to have several vulnerabilities, some of
these vulnerabilities will result in greater damage than others. Some these
vulnerabilities may “contribute significantly to the enemy’s downfall while others may
lead to only minimal gains” [1]. Those vulnerabilities which offer the greatest impact
are known as critical vulnerabilities. These are the vulnerabilities that are to be pursued
by attacking forces and should be the focus of efforts.
The ubiquity and prevalence of information systems increases the overall attack
surface and number of exposed vulnerabilities. Finding, classifying, and determining
which of these vulnerabilities are “critical vulnerabilities” is crucial in the effective
employment of offensive cyber capabilities. Once again, planners must not silo cyber
capabilities, as information weapons can be used to create opportunities for maneuver
against conventional critical vulnerabilities and ultimately CoG. Much effort has
already been placed in determining the CoG and critical vulnerabilities in planning for
conventional warfare, these CoG should be reevaluated to find avenues where offensive
cyber capabilities can help maneuver against critical vulnerabilities and create
opportunities for attacks against CoG. Critical infrastructure is one such conventional
CoG that has already been identified where offensive cyber capabilities can bring new
avenues of attack and exploitation [15]. The enemy’s critical infrastructure has always
been a prime target for conventional and kinetic weapons, information weapons simply
bring new avenues to reaching and disrupting this critical infrastructure. As planners
begin to understand the capabilities of offensive information weapons, some CoGs and
critical vulnerabilities will change. Not all of these new CoGs and critical vulnerabilies
will be purely military, civilian infrastructure will be affected as well. As CoG to blur
from military objectives to civilian objectives, war will truly become an “extension of
politics” as opposed to a struggle between two different military forces.
2. New Weapon Systems, Classic Principles
“Over time, perhaps as little as in twenty years, and as the leverage provided by
technology increases, this threshold will finally reach its culmination - with the ability
of one man to declare war on the world and win."
John Robb, Brave New War
Conventional weapons (rifles, indirect fire weapons, explosives…etc) have physical

limitations. These physical limitations cannot be overcome even if the individual
employing that weapon system is highly skilled or experienced with that weapon
system. These physical characteristics provide the basis for the development of tactical
guidance for effective employment of the weapon system. For example, the M4 carbine
is designed to be employed as a short/medium range weapon in force while the M24
Sniper Weapon System is designed for precision, long range fire from trained
marksmen [16]. This foundation for the employment guidance for the M4 and M24 are
based on the specific characteristics and physical capabilities of the weapon system.
The skillset of the individual employing the conventional weapon system may stretch
the weapon systems capabilities in a small way, but the ultimate capabilities of the
weapon system remained tied to the physical characteristics of the weapon.
The weapons (information weapons) used in cyber-attacks are different from
conventional weapons. With conventional weapons, the physical weapon system
represents the capabilities being brought to the battlefield. Information weapons enjoy
a different type of relationship. With information weapons, the attacking power of the
weapon system is directly correlated to the skillset of the individual using the weapon
system. The capabilities of conventional weapons systems are bound to the physical
characteristics of the weapon system. Two identical rifles fired by two differently
skilled operators will continue to fire with the same muzzle velocity and rate, as the
physical characteristics constrain the ultimate capabilities of the weapon. Information

weapons on the other hand, are bound directly to the skillset of the individual
employing the information weapon [17]. Two identical laptops employed by two
differently skilled operators will have completely different capabilities. As the
individual’s skillset increases, so does the striking power and effectiveness of the
information weapons employed by that individual. Creating an offensive cyber
capability is less about finding the right hardware and more about finding the right
people and skillsets.
The importance of the individual brings about unique challenges for intelligence
organizations when attempting to understand and estimate the enemy’s offensive cyber
capabilities. With conventional weapon systems, capability can be tracked via
procurement, tracking of the physical location of the weapon systems, and active
surveillance of the weapon system. Physical movement and logistical operations
associated with the employment of conventional weapons are eagerly watched by
intelligence organizations and is used in all types of intelligence analysis. Developing a
nuclear capability for example, require distinctive materials, specialized knowledge,
and distinctive facilities for development. Non-proliferation is tracked through
safeguards which monitor materials, inspections of facilities, and surveillance. Each
safeguard has a threshold that indicates when a nation state may be attempting to create
a nuclear capability. Nation states developing an offensive cyber capability can do so in
a much more subtle way. There are no specific thresholds, distinctive materials, or
facilities that indicate an offensive capability is being developed [18]. Individuals with
a solid understanding of offensive security can be trained or recruited from both
academia and corporate environments putting impressive offensive cyber capabilities
within reach of every nation, regardless of size or economy. These individuals are the
capability. With information weapons, the capability rests with the operator of the
information weapon, not the equipment itself. The commercially available laptop
available at any major retail outlet can be used to conduct attacks against any nation in
the world. The striking power of this attack is measured not by the hardware, but the
skillset of the operator. This makes tracking via procurement and logistical operations
impossible. The wide spread availability of sufficient hardware coupled with the lack
of distinctive, easily tracked characteristics not only lowers the barrier for entry for
establishing an offensive cyber capability, it makes determining the true source of the
attacks virtually impossible. Intelligence organizations must now shift focus from
identifying physical equipment and logistical actions to identifying key capabilities and
specific skillsets possessed by individuals. This is an extremely difficult and daunting
task, making determining the true capabilities of nation’s offensive cyber capabilities
difficult.
In August of 2008, the Grey Goose project kept a Russian hacker forum under
surveillance watching the interaction of the various forum members. Investigators
determined that the forum had over 600 registered members (users were required to
register in order to read/write posts). A sample of the forums member list is shown in
figure 5.
It was impossible to immediately determine which of the forum participants
represent a legitimate offensive capability and which forum members are simply “script
kiddies” (unskilled participants). Analysts for the Grey Goose Project were forced to
analyze thousands of forum events, learning about the various topics being discussed
by the forum members. Each forum post was analyzed for technical sophistication and
technical leadership. Relationships between forum members were mapped using
Figure 5. Member List from a Russian Hacker Forum
technically sophisticated Palantir analysis platforms [19]. Only after extensive analysis
could the Grey Goose investigators determine which members represented the true
offensive capability of the forum. Once these individuals were identified, surveillance
focused was focused on these key individuals. Each individual was noted and ranked
using forum participation as the key indicator of their technical sophistication
(individual contributed vulnerabilities, contributed tools, provided advice for
exploitation…etc.). This approach allowed the analysts to focus on the handful of
individuals driving the offensive capabilities of the entire forum. It was these
individuals that were offensive capability, not the tools, hardware, or even the forum
[4].
3. Conclusions
“Preparing to win in combat must be the highest priority in the allocation of time,
dollars, and rewards, at every level and under all circumstances”
William S. Lind, Maneuver Warfare Handbook
3.1. Employment
The focus on using cyber capabilities to “win wars” must be at the forefront when
developing doctrine. It is easy to become enamored with the seemingly magical
displays of exploitation and technical jargon; however planners must recognize that
cyber capabilities represent one of many dimensions of warfare. Planners must not silo
cyber capabilities, employing them in isolation, but must consider how best to augment
conventional capabilities with cyber capabilities. Having a robust cyber capability is
important, however winning the “cyber-battle” while losing the conventional war is
unacceptable in any scenario. In time, information weapons will be the weapons of
force, perhaps establishing themselves as the “main effort” in campaigns with
conventional forces designated as “supporting efforts”. Until that time, information
weapons are to be employed much like other conventional weapon systems as
supporting efforts, helping shape the battlefield in support of the main effort (typically
conventional forces). Planners must strive to integrate cyber capabilities into
conventional warfare as a supporting arm. Information weapons, much like other
supporting arms, must be cognizant of the main effort and should strive to shape the
battlefield in support of the main effort. Commanders must be acclimated as to how to
request supporting cyber capabilities and understand what gaps offensive cyber
capabilities can cover.
3.2. Command
Rigid, highly centralized command makes the development and effective employment
of offensive cyber capabilities difficult. Commanders must be careful not to impose
rigid requirements or artificial constraints onto cyber capabilities. Judicious use of
commander’s intent is essential in establishing the decentralized operational command
necessary for the development and effective employment of offensive cyber
capabilities. A top down, micromanaged effort will kill the speed, tempo, and most
importantly the creativity required in effective cyber-attacks. Hierarchical, centralized
administrative chains of command are essential for the good order and discipline in
military ranks; however operational chains of command should strive to push decision
making down to the lowest level, using commanders intent to guide decision making
and initiative. This forces a more decentralized approach to employment of cyber
capabilities, allowing for the need flexibility needed to successfully employ offensive
cyber capabilities. Without this decentralized approach, employment of offensive cyber
capabilities will ultimately fail.
3.3. The Individual is the Weapon System
The leverage technology brings coupled with the increasing ubiquity of information
systems builds upon the power wielded by individuals with the right skillsets. As
operational commands become more and more decentralized and the impact of the
individual becomes more and more powerful eventually, a tipping point will be reached
and the individual will represent the offensive capability. The concept of the measuring
a nation state’s striking power and capability through the surveillance and tracking of
ground forces, air, and naval equipment will eventually succumb to the identification of
highly skilled individuals that represent the offensive cyber capability. As our reliance
on technology continues to evolve, the ability of the lone individual to disrupt
conventional operations also increases. Eventually, the power of the lone individual
will grow until a lone, highly skilled individual (or a small team of highly skilled
individuals) will be able to impose their “political will” on other individuals,
corporations, and even nation states.
References
[1] United States Marine Corps, Marine Corps Doctrinal Publication 1 - Warfighting, United States
Marine Corps, Quantico, VA, 1997. http://www.dtic.mil/doctrine/jel/service_pubs/mcdp1.pdf
[2] Department of Defense, Conduct of the Persian Gulf Conflict, Department of Defense,
Washington, DC, 1991. http://www.dod.gov/pubs/foi/reading_room/305.pdf
[3] Correlli Barnett, Victory Through Air Attack? It’s a Pie in the Sky. TimesOnline, 2009.
http://www.timesonline.co.uk/tol/comment/columnists/guest_contributors/article5469051.ece
[4] Jeffery Carr, Grey Goose Phase I. GreyLogic, 2008. http://www.scribd.com/doc/6967393/Project-Grey-
Goose-Phase-I-Report
[5] Kevin Benson, Tactics for Small Wars. Institute of Land Warfare, 2008.
http://www.ausa.org/SiteCollectionDocuments/ILW%20Web-
ExclusivePubs/Landpower%20Essays/LPE08-1.pdf
[6] Marine Corps Combat Development Command, Offensive Fundamentals I. Marine Corps Combat
Development Command, 2009. http://www.usna.edu/USMCInfo/Documents/Pubs/b0354.pdf
[7] Lt Col. Lawrence Shattuck, Communicating Intent and Imparting Presence, Military Review, 2000.
http://www.au.af.mil/au/awc/awcgate/milreview/shattuck.pdf
[8] Ives, Walsh, Schnieder, The Domino Effect of Password Reuse, Communications of the ACM, 2004.
http://portal.acm.org/ft_gateway.cfm?id=975820&type=pdf&coll=GUIDE&dl=GUIDE&CFID=38680
500&CFTOKEN=56506836
[9] Kenneth Corbin, Lessons From The Russia-Georgia Cyberwar, Institute of Communications Studies,
2009. http://ics.leeds.ac.uk/papers/vp01.cfm?outfit=gdr&folder=442&paper=750
[10] Whitaker, Evans, & Voth, Chained Exploits: Advanced Hacking Attacks from Start to Finish, Pearson,
2009.
[11] Office of the Secretary of Defense, Military Power of the People’s Republic of China 2006, Office of
the Secretary of Defense, 2006. http://www.defenselink.mil/pubs/pdfs/China%20Report%202006.pdf
[12] William Jackson, US Already at War in Cyberspace, Government Computer News, 2009.
http://www.gcn.com/Articles/2009/04/22/RSA-cyberwar.aspx
[13] Toomas Hobermagi, Estonia Promises Georgia Help in Fighting a Cyberwar, BalticBusinessNews.com,
2008. http://balticbusinessnews.com/Default2.aspx?ArticleID=4743f52b-6f71-4ebd-ad1c-
ca4ab13ad921
[14] Wikipedia, OODA Loop, Wikipedia.com, 2009. http://en.wikipedia.org/wiki/OODA_Loop
[15] Cordesman, Cyber-Threats, Information Warfare, and Critical Infrastructure Protection, Greenwood
Publishing 2002.
[16] Department of Defense, Basis of Issue Plan – M24, Department of Defense.
http://www.fas.org/man/dod-101/sys/land/docs/bnI063AA.htm
[17] Killgannon & Cohen, Cadets Trade the Trenches for Firewalls, Nytimes.com, 2009.
http://www.nytimes.com/2009/05/11/technology/11cybergames.html?_r=2&ref=technology
[18] Wikipedia, Nuclear Proliferation, Wikipedia.com, 2009.
http://en.wikipedia.org/wiki/Nuclear_proliferation
[19] Palantir Technologies, Palantir Government Services, 2009. http://www.palantirtech.com/
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-156
Belarus in the
Context of European Cyber Security
Fyodor Pavlyuchenkoa
Translated from the Russian language by Kenneth Geersb
a
www.charter97.org
b
Cooperative Cyber Defence Centre of Excellence
Abstract. During the first decade of the 21st century, Internet censorship in
Belarus has evolved into a government tool used to combat political dissent. State-
sponsored denial of service (DoS) attacks against civil society have become a
domestic crisis that threatens not only freedom of expression in Belarus, but also
the integrity of Internet resources throughout Europe. The ongoing cyber conflict
between state and non-state actors in Belarus is analogous to the struggle between
the Russian government and its internal adversaries in cyberspace. In this essay,
we recount the history of cyber censorship and attacks against Charter ’97, a
popular Belarusian website, and discuss the effectiveness of countermeasures.
Keywords. Belarus, Internet, censorship, charter ‘97, denial of service, cyber

crime
Introduction
For over a decade, the Charter ‘97 website in Belarus has been a leading venue for
Belarusian public policy discussion. Since its founding, however, the site has been
forced to defend itself against practically all existing Internet censorship strategies.
Following a disputed political referendum in Belarus in 1996, Alexander Lukashenko
alone has governed the country. His government has suppressed freedom of speech,
and Charter ‘97 is well-known for siding with Belarusian dissidents.
While modern technology has offered the world significantly improved
communications, it also creates new threat vectors. Nation-states can abuse their power
over telecommunications, creating dangerous precedents and fostering long-term
political instability.
In cyberspace, the activities of the Belarusian secret services are reminiscent of
their colleagues in Russia. It is critical that democratic states in Europe strengthen
independent Internet resources in all European countries, and strive to extend the rule
of law in the whole of European cyberspace.
1. Background: Internet Censorship in Belarus
For over a decade, there has been virtually no independent traditional media in Belarus.
Under pressure from the authorities, the popular newspapers of the 1990’s have ceased
to exist or have seen their circulation greatly reduced. The independent FM radio
F. Pavlyuchenko / Belarus in the Context of European Cyber Security 157
stations have been closed, and an independent Belarusian television channel has never
existed. It is therefore unsurprising that, despite its high cost, the Internet has been and
remains the only source of objective information for the majority of Belarusians, and
the number of Web users has grown to nearly one-quarter of the country’s population.
09.09.2001
On September 9, 2001, at 1200, Belarusian Internet users came into conflict with the
authorities for the first time – on the day of the national presidential elections. The
national telecommunications firm Beltelecom – a monopoly provider in the country –
intentionally blocked access to a range of popular political websites. By the following
afternoon at 1600, all Internet censorship had ceased.
From a technical perspective, such information blocking is easy for a
telecommunications monopoly to perform. The data packets can be filtered on the
ISP’s primary router based on their source or destination Internet Protocol (IP)
addresses. With the help of the ‘tracert’ (traceroute) command, the point of network
interruption was easy enough to find. During this event, the prohibited sites continued
to be accessible outside Belarus, but not within the country.
Some of the popular but blocked sites, including *.home.by, *.minsk.by, *.org.by,
*.unibel.by, *.nsys.by, and *.bdg.by were physically hosted on servers in Belarus,
within the .by network domain. These sites were disabled by altering their Domain
Name Service (DNS) records. In that context, it is important to note that the
management of the .by Top Level Domain (TLD) is the responsibility of a special state
agency, the Operations and Analysis Center, which falls under the direct control of the
President of Belarus.
Some websites, including www.charter97.org, created multiple mirror sites in an
effort to stay online. All such mirrors were promptly blocked by the government.
Furthermore, popular ‘anonymizer’ websites and pages that published lists of free
proxy servers were also blocked. In all (including the sites that were directly and
indirectly blocked), over 100 websites were inaccessible.
It must be emphasized that there were no legal grounds to perform Internet
censorship in this case. In fact, such censorship directly violated the Belarusian
constitution. The official explanation from the Ministry of Communications and
Beltelecom was that too many Belarusians were trying to access the sites in question at
one time, and that this led to a self-inflicted denial of service. From a technical point of
view, that explanation does not hold water. For its part, the Belarusian government had
no comment, even though Internet censorship – in this case, computer sabotage – is an
offense under Belarusian law. An official investigation into the facts of the case was
never undertaken.
24.10.2001
The Charter ’97 website was completely deleted from its server by unidentified
hackers. A few days after this incident, under pressure from the Belarusian secret
services, the hosting company was forced to break its contract with Charter ’97, and the
site was no longer allowed space on its server.
158 F. Pavlyuchenko / Belarus in the Context of European Cyber Security
20.01.2004
In January 2004, Charter ’97 was the target of a massive distributed denial of service
(DDoS)-attack for the first time, which lasted more than 3 weeks. The attack followed
the publication of a journalistic investigation into a possible connection between high-
ranking officials from the Belarusian Interior Ministry (from a department responsible
for investigating computer crimes) and the trading of child pornography on the Internet.
In a strange coincidence, Natalya Kolyada, a human rights activist working with our
website at that time, was also convicted on misdemeanor charges.
The DDoS attack was supported by a botnet that included more than 55 thousand
active IP addresses. This network of infected computers spanned the globe, and
included compromised machines in Latin America, the United States, South-East Asia,
China and India. The geographic dispersion of the botnet allowed the power of its
attack to be spread across a 24-hour period. Further, the power and focus of the attack
changed several times, which indicated an active command and control (C2) over the
activity.
While it is impossible to affirm that this attack was politically motivated, a
simultaneous campaign of harassment was organized against the employees of our site
on official Belarusian television. Our employees were, among other things, accused of
trading in online pornography.
14.07.2004, 21.07.2004
On July 21, 2004, there were mass protests in Minsk to mark the 10th anniversary of the
Lukashenko government, and our website again came under a DDoS-attack. Charter
’97 had planned to host a webcast to cover the protests. One week prior, on July 14, a
‘test’ cyber attack had paralyzed our server for 2 hours. On July 21, the main attack
began at 1400 – 4 hours before the demonstrations began – and lasted until the political
protests were over. The technology and power of the attack were similar to the attack in
January of that year.
10.10.2004
The next large-scale attempt to block Charter ’97 and other independent websites
occurred in the autumn of 2004, during parliamentary elections and a national
referendum on the lifting of presidential term limits in Belarus. On the day before the
election, correspondents were unable to access our website and could not call us by
mobile or landline telephone. At the same time, various opposition websites were again
blocked by a filter on Beltelecom’s primary router.
This time, many Belarusian network users were better prepared to combat
censorship, and immediately switched to Internet proxies and anonymizers. But the
Belarusian Internet authorities had an effective new weapon in their arsenal: the
artificial stricture – or “shaping” – of Internet bandwidth. The use of this tactic meant
that, in principle, forbidden sites were still available. However, it took anywhere from 5
to 10 minutes for the censored webpages to load in a browser. Thus, most Web users
were unable to gain full access to Charter ’97, as well as a range of other targeted sites.
All other Internet resources were accessible as normal.
There was no announcement from the Ministry of Communications or Beltelecom
regarding this incident, nor was any official investigation undertaken.
19.03.2006
The next time that websites were blocked in Belarus was on March 19, 2006 – the day
of Belarusian presidential elections. Well before the election took place, in an initiative
called ‘Free Internet’, Charter ‘97 began to offer site visitors information regarding
various ways to circumvent censorship. This strategy ensured that all attempts to block
information using IP-filtering failed. However, network ‘shaping’ – or the intentional
stricture of specific streams of network bandwidth – was again the method employed.
On March 18, the day before the election, a website filtering ‘test’ was conducted
from 1600-1630. On election day, sites belonging to opposition presidential candidates
and their political party websites, leading independent news sources, and
www.livejournal.com (an international blogging site very popular with Belarusians)
were blocked.
Representatives from Beltelecom announced that the technical interruptions were
caused by the overloading of particular circuits, and no formal investigation was ever
undertaken.
25.04.2008
On the eve of massive street protests in Minsk, the government changed its strategy. At
1420, a test DDoS-attack on Charter ’97 lasted 30 minutes. The following IP addresses
were used in the attack: 89.211.3.3, 122.169.49.85, 84.228.92.1, 80.230.222.107,
212.34.43.10, 81.225.38.110, 62.215.154.167, and 62.215.117.15. For about 10
minutes, our site was difficult to access, but we were able to restore normal traffic
before the attack ended.
On April 26, the main DDoS-attack took place. It began five hours before the start
of the demonstration. Charter ’97 had intended to conduct a live webcast of the
protests, but the attack paralyzed our server. Our hosting company,
www.theplanet.com, attempted unsuccessfully to mitigate the attack. Its hardware was
designed to defend against an attack only up to 700 Mbps. The volume of this DDoS
reached over 1 Gbps. We had no choice but to turn off the site and simply wait for the
attack to end. The perpetrators were apparently satisfied with their achievement, and
the attack was over by the next day. It is important to note that other independent
online media were subjected to similar attacks at the same time, especially ‘Belarusian
Partisan’ and the Belarusian-language version of ‘Radio Liberty’.
The server hosting ‘Belarusian Partisan’ crashed as a result of the DDoS-attack.
Further, system administrators even temporarily lost control of the site, and for several
days, unknown hackers used the website to publish fabricated, scandalous news stories.
The Belarusian Partisan editors were forced to denounce the false reports on other
websites. The level of expertise required for this attack was high enough that there is
no doubt the Belarusian special services were involved in the incident.
The technical capabilities of the Radio Liberty (RL) server – home to the
Belarusian, Albanian, Azerbaijani, Tajik, and Russian RL services – were sufficient to
contain a similar attack for more than 3 days. However, the site was nonetheless
difficult to access until 1500 on April 28, and that was enough to cause a minor
diplomatic scandal. The U.S. mission to the Organization for Security and Cooperation
in Europe (OSCE) issued a statement on the cyber attack. The Belarusian Ministry of
Foreign Affairs publicly denied any official involvement.
Figure 1. Explanatory email from the Internet Service Provider (ISP)
Increasingly, experts believe that various political DDoS attacks share some
common characteristics, and that there may be important links between discrete attacks
such as those which hit Estonian and Georgian websites.
08.06.2009
The most recent example of a politically-motivated DDoS attack on Charter ’97

occurred in June 2009. The incident may have been related to the recent conflict
between the governments of Russia and Belarus, which resulted in the imposition of
economic sanctions against Belarus and a worsening political situation inside the
country.
The DDoS attack lasted more than a week, and for a while it paralyzed our site
completely. The strength of the DDoS-attack in this case had not been particularly
high; about five thousand IP addresses took part in it. With the support of our ISP, the
Charter ‘97 technical support staff was able to neutralize the attack.
Figure 2. Timeline for the June 2009 attack against Charter ‘97
2. Countermeasures and their Effectiveness
Throughout the history of these DoS attacks, we have been looking for ways to counter
censorship by the Belarusian authorities. We have tried many different methods, but
none of them has been completely effective. Our situation might best be described as a
competition to outmaneuver an opponent who has more resources than we do.
Initially, our strategy was to strengthen our technical capabilities and to increase
our expertise in computer security. The site was moved to a relatively powerful,
hardened server, and we created a system for monitoring vulnerabilities and attempted
intrusions. We employed encryption, not only for access to the server itself, but also for
access to our content management system. We created a multi-tiered system of access
to the server and to the site, as well as the ability to quickly replace all passwords in the
event administrators and/or journalists were arrested. We developed a distributed
system for creating server data backups. Based on our experience so far, it is best to use
simple, open-source technologies such as UNIX, PHP, and MySQL to help with site
mobility (i.e. the rapid transfer of the site to another hosting platform). Firewall and
caching technologies are sufficient to repulse DDoS-attacks of average strength.
Combined, these efforts helped to prevent subsequent compromises.
Separately, Charter ’97 launched the ‘Free Internet’ project. This website provides
recommendations to site visitors for what to do in case the site is blocked, and explains
how to use an Internet proxy, anonymizer, Virtual Private Network (VPN), and
software such as Tor. Visitors are encouraged to disseminate information
independently through their own blogs, forums, chat rooms, social networking sites, e-
mail, and instant messengers. Site information is rebroadcast, for example, via RSS or
email to mirror sites and partners. The successful use of these measures can overcome
the simple blocking of IP addresses. However, with our limited resources, there is no
current, solid countermeasure to DDoS.
We believe that the authorities have concluded that their most effective methods of
censorship are DoS attacks and information manipulation. In support of the latter,
specially-trained ‘visitors’ insert themselves into discussions on popular websites, and
monitor or help ‘guide’ the course of discussion. During politically sensitive times, or
whenever political dialogue becomes excessive, the site can be temporarily blocked by
DDoS.
3. Government Power and Cyber Crime
The current power structure in Belarus not only attempts to suppress political dissent
on the Internet, but also flagrantly violates the Belarusian constitution. In Belarus, there
is no legal basis for Internet censorship, much less for computer hacking and DoS
attacks on websites. There is only one case of censorship in Belarus that was officially
acknowledged and at least somewhat justified: in 2005, Beltelecom blocked two
Russian gay sites for possession of pornography, at the behest of the Ministry of
Culture’s Republican Commission for the Prevention of Promoting Pornography,
Violence and Cruelty. All other cases of Internet censorship involved the use of
blatantly criminal methods.
DDoS-attack techniques against Belarusian independent Internet media could
easily be used to block all sites covering current affairs in Belarus. Further, the lack of
the rule of law within Belarusian Internet space could facilitate the growth of organized
cyber crime on the international level.
There is active cooperation between Belarusian and Russian special services in
cyberspace: the Agreement on Cooperation of the Commonwealth of Independent
States (CIS) in Combating Cybercrime was signed in 2000. Finally, there are
similarities in the cyber attack methods used against Estonia, Georgia, and the websites
of human rights organizations in Belarus and Russia, which suggest that these crimes
have common roots.
4. What is Required?
The challenge of DDoS attacks threatens civil society throughout Eastern Europe. In
Belarus, Ukraine, Russia, Georgia, Armenia, and Azerbaijan, the government may
already have used DDoS attacks as a tool for countering dissent on the Internet. To
mitigate this threat, an international, collaborative approach is required. A good start
would be the establishment of an international hosting platform designed to support
freedom of speech throughout Europe, built by a team of international experts. They
should investigate cyber crimes based on aggregate data, and work toward the
development of effective defense methods and technologies. The mere creation of such
a platform would be a helpful step, and could enhance the level of cyber security and
freedom of expression throughout Europe.
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-163
Politically Motivated
Denial of Service Attacks
Jose NAZARIO
Arbor Networks, United States
Abstract. Cyberwarfare has been waged for well over a decade, utilizing methods
such as website defacement, data leakage, and distributed denial of service attacks
(DDoS). This paper focuses on the latter, attacks that are easily carried out and
designed to overwhelm a victim’s network with wasted traffic. The goal of a
DDoS attack is to make the use of the network impossible for internal or external
users. Through a brief examination of the history of these attacks, we find they
previously were designed to inflict punitive damage on the victim but have since
grown into sophisticated censorship tools. Our approach measure such attacks by
looking at Internet backbone traffic, botnet activities, BGP routing changes, and
community chatter about such attacks to provide a robust picture of politically
targeted DDoS attacks. Our analysis indicates that most of the attackers are non-
state actors but are able to fluidly utilize a growing botnet population to launch
massive denial of service attacks. This finding has broad ramifications for the
future of these attacks.
Keywords. DDoS, botnets, Estonia, Georgia, China, Russia
Introduction
Internet attacks take on many forms, including system compromises and information
theft, as well as denial of service attacks designed to disrupt services. Motivations for
cyber-attacks include frustration, fun, and extortion, especially against gambling and
pornography sites. Anger and frustrations appears to be the major motivation in attacks
against gaming sites and forums, where player-on-player attacks happen quite
frequently. Politically targeted attacks are extremely rare in the sphere of daily attacks.
The types of attacks launched depend on the attackers’ skills and motivations.
A distributed denial of service (DDoS) attack is nothing more than a coordinated
effort that instructs PCs to send a victim a flood of traffic designed to overwhelm their
servers or consume their bandwidth [[1]]. Regardless of the attacker’s underlying
motivations, the attacks are designed to disrupt the normal flow of the site for internal
or external users. The PCs used in the attacks can be the bots in a botnet or a zombie
army, or it can be tools willingly installed on peoples’ computers. A simple form of a
DDoS attack is when individuals work together and to continuously reload a website in
a browser such as Internet Explorer. In each case, the purpose is to aggregate the PCs’
bandwidth together to overwhelm an adversary who is usually superior in their
bandwidth resources, and to do so from a large enough number of locations to make
source-based filtering unmanageable.
164 J. Nazario / Politically Motivated Denial of Service Attacks
DDoS attacks are among the most visible and disruptive of cyber-attacks. When
coupled to political motivations, they can be seen as an extension of politics in the 21st
century, to borrow a phrase from von Clausewitz. Currently, researchers infer a
political motivation for various attacks based on internal information, such as the
nature of the victim and the attack commands seen. Investigators may also use external
sources to validate this finding by looking at news reports and website conversations
discussing diplomatic grievances and their redress through online attacks. In this
context, cyber-attacks are sometimes referred to as “fifth generation warfare”.
Arbor Networks’ Peakflow products are used by many Internet service providers to
detect and defend against DDoS attacks [[2]]. Independently run Peakflow
deployments collect data on such attacks and provide a distillation of the events to the
Arbor Networks ATLAS portal. Attack data is gathered in three different ways to
provide a nearly complete picture.
The first data source is direct traffic measurement using Arbor Peakflow
deployments around the world using data collected as part of the ATLAS project.
Customers can share attack information with each other through the Fingerprint
Sharing Alliance. Some of this data is made available in ATLAS and can be analyzed
by network or country affected or launching the attack. Peakflow counts attacks based
on the traffic types (e.g. TCP SYN, ICMP echo request) and destination networks for a
time period using dynamically learned baselines or static thresholds. Therefore,
Peakflow may register multiple concurrent attacks if they target the same destination
but use different traffic, such as a simultaneous TCP SYN flood and ping flood.
The second way attacks are measured is to look at commands sent to botnets to
launch attacks. Malicious software analysis can be used to discover botnets and
infiltrate them by communicating with the botnet’s command and control (C&C) server
by mimicking legitimate bot clients, enabling a record of the botnet’s activities for later
analysis. This data is valuable to understand the attack’s root origins for disruption but
also for post-event analysis to understand the nature of the attackers. Most of the
attacks tracked are against inconsequential targets, but sometimes they target victims
such as financial firms, major e-commerce sites, or government assets.
The third form of continuous measurements is to look at border gateway protocol
(BGP) routing data used to provide Internet backbone routing. Sometimes the paths
may change during an attack as a direct result of the attack, such as BGP session drops
during congestion, or through attempts to mitigate the attack. Changes to the BGP
routes for a victim can indicate an attack.
This paper focuses on the confluence of DDoS attacks with political targets and
political or ideological motivations. DDoS attacks are crippling because they are
designed to make the networks they target unusable, either to inflict damage to the
victim or, in the case of many recent events, to silence their opponents by making their
resources inaccessible. This paper does not analyze information on attacks such as
website defacements or compromises through malicious software that may be a part of
these attacks.
1. Major Events in Political DDoS Attacks
DDoS attacks became widely popular in the late 1990s following the development of
toolkits such as Tribe Flood Network and Trinoo [[1]]. These methods were quickly
J. Nazario / Politically Motivated Denial of Service Attacks 165
adapted for political targets. Major attacks from the past 10 years can be used to
highlight changes and illustrate how sweeping this problem can be.
Very early events in this field include attacks on NATO computers in the former
Yugoslavia during the campaigns in the late 1990’s, and also the attacks from Chinese
hackers on US military sites following the bombing of the Chinese embassy by a US
plane in the former Yugoslavia during that NATO mission [[3]]. This list of attacks
shows how many different regions are affected and how many different motivations
exist for these attacks. It also shows how these attacks have evolved over time. This
section also shows that such attacks didn’t start with Estonia in May 2007 and didn’t
end with Georgia in the summer of 2008.
1.1. Hainan Spy Plane Incident
In April 2001, a US Navy spy plane was on a reconnaissance mission off the southern
coast of China when multiple Chinese fighter planes intercepted it, and one of the
planes clipped the US Navy plane, causing damage to both planes. The pilot of that
Chinese fighter plane was lost after his plane broke up in mid air, and due to the
damage it sustained during the accident, the US Navy plane had to make an emergency
landing in Chinese territory on Hainan Island. The crew was held for several days
before diplomatic efforts released them.
During this time, tensions between the US and China ran high. Among the events
that occurred were multiple attacks, including DDoS attacks and probes on US military
Internet sites. The Chinese hacking group “Honker Union” is believed to have been
behind the attacks [[4]].
The attackers in this situation appeared to see these attacks as acts of patriotism.
The public outrage was undeniable and bubbled over to Internet forums. Multiple
groups and parties appeared to take part in these actions.
1.2. Estonia, 2007
Beginning in late April 2007, the European nation of Estonia was hit by a series of
coordinated denial of service attacks. Ethnic Russians make up a significant percentage
of Estonia’s population, and by many accounts Estonians and the ethnic Russians co-
existed peacefully [[5]]. As is commonly found throughout Russia and much of the
former Soviet Union, Estonia has a statue of a Soviet soldier commemorating the end
of World War II.
The statue has been a sore point in Estonian politics for many years and was
moved in April 2007, leading to civil unrest within Estonia and complaints by the
government in Russia [[14]]. Coinciding with the street protests, online DDoS attacks
began to target Estonian government and private sector sites, including banking
institutions and news sites.
The attacks seen in Estonia built up over the course of a few weeks and peaked on
Victory Day, May 9. On this day, Peakflow systems around the world measured attacks
lasting 10 hours each with a peak bandwidth utilization of 95 Mbps. This data comes
from multiple Peakflow sensor sources that are aggregated into ATLAS via ISPs that
provide transit for Estonian ISPs [[6]].
The attackers used multiple attack methods. They used Russian language forums
and blogs to spread tools such as ping flood scripts and to coordinate their efforts, and
they also recruited botnets into the effort. For example, they worked hard to take their
collective tools, botnets, and activities and fire them at the same time (e.g. 11pm in
Moscow).
The attacks in Estonia hit many parts of the infrastructure, including the websites
for the prime minister, parliament, various ministries, and even government name and
mail servers. News reports contained information about slowdowns with some banks
and financial transactions. All of this is consistent with a nation that makes heavy use
of the Internet for daily life suffering from systemic flooding.
Most of the attacks measured in ATLAS died out after Victory Day, although
reports from first-hand accounts within Estonia indicate that they continued for several
weeks.
1.3. China and CNN
In April 2008, the CNN news personality Jim Cafferty commented on air about the
Chinese preparations for the Olympics in Beijing, China. Many Chinese found these
remarks offensive, and this sentiment quickly brewed into anti-CNN hacking events.
A number of hacking groups activated and worked to coordinate their activities.
The attacks included website defacements and many probes to try and disrupt the
CNN.com website. Peakflow and ATLAS also monitored the flows for the site as well
as for botnet attack activity [[7]].
During the investigations, a number of Windows tools developed to target CNN
specifically were discovered, in addition to a few botnets that were targeting CNN
more generally. The first tool was dubbed “Supper DDoS” is a simple flooder usable
by an average computer user, with only an input for the victim’s address together with
“Attack” and “Stop” buttons. This tool was distributed on Chinese language forums by
an unknown number of authors.
Researchers also discovered a botnet apparently operated by “Ice Kernel”, using a
bot dubbed “KernelBot”. KernelBot is a flexible DDoS attack system, supporting
common attack types, as well as full control of the victim’s PC. Commands for this
botnet targeting CNN’s website appeared during this event. Another tool released in
late April 2008 to target CNN was a specialized version of the NetBot Attacker tool, a
general purpose DDoS tool that is usually deployed on a victim’s PC using standard
malware infection methods. This particular version of NetBot Attacker is hard coded to
target CNN.com and provides the user with some basic control over their PC. This kit
includes the flooding portion of the bot and the attacker’s UI for control, something not
normally seen. Typically, the bots run without any UI for the victim.
1.4. Georgia and Russia
In July 2008, the website for Georgian President Mikheil Saakashvili was hit with a
DDoS attack. In this case the botnet was based on a codebase that is only seen in
Russian-language botnets. The command and control server for this botnet was located
in a regional ISP, PaeTech, and had been under surveillance by ATLAS and other
researchers for some time. This was the only attack launched by this botnet and lasted
from July 18-20, 2008 [[9]].
These attacks were corroborated together with ShadowServer, a volunteer botnet
monitoring team. We attempted to reach the site during the attack and found that the
Georgian President’s website was unable to load from a number of North American
vantage points, consistent with a major attack [[10]]. When asked by the press,
spokespeople for Saakashvili’s office said that no such attacks had occurred, however.
A message was included in the attacks that read “win love in Russia”, consistent
with the ongoing tensions in the region. A few days before the DDoS attacks began in
July 2008, the ITAR-TASS news agency from Moscow ran a story with the translated
headline “Withdrawal of Georgian troops only way out of Abkhazia conflict -
Medvedev”. At the time, Russian president Dimitry Medvedev had been in power for a
few months. There had been ongoing, minor skirmishes between Georgia and Russia
over two regions within Georgia. South Ossetia and Abkhazia, two semi-autonomous
areas have historically stronger ties to Moscow than does the rest of Georgia. These
two regions had been seeking more independence and closer ties to Russia than Tbilisi
would allow. The diplomatic flames going back and forth were substantial and included
reports of gunfire between Georgian and Russian forces. After the shutdown of the
PaeTech C&C server, the July 2008 attacks stopped [[9]].
A few weeks later, in early August, a large-scale shooting war between Georgia
and Russia broke out with Russian tanks entering Georgian territory. Almost
immediately, very substantial DDoS attacks began to flood into Georgia and were
caused by multiple botnets and ping flood scripts. Targets included the Georgian
president’s site, various ministries, news agencies, and others [[6]]. Furthermore,
ATLAS monitors recorded some attack commands into Russia at the same time,
suggesting that someone - either Georgian or possibly a Georgian sympathizer - tried to
counter attack.
Arbor Peakflow and ATLAS live traffic monitors on the Internet showed that the
peak size of the attack was substantially larger than the attacks in Estonia the year
before. The peak bandwidth recorded during the attacks was over 800 Mbps, and the
attack were much more intense [[9]].
Key Georgian properties were quickly relocated to various countries with better
defense capabilities. The president’s website, for example, was moved to Atlanta
Georgia and Tulip Networks. Other sites were moved to Estonia, which had experience
and tools after the previous year’s attacks [[11]].
This was the first time in nearly 10 years that a military conflict and a cyber
conflict coincided, the most recent being attacks between Israel and Palestinian
militias. These attacks on Georgian websites, especially after what happened in
Estonia, have raised concerns around the world by governments concerned about an
apparently growing trend of politically motivated attacks on government networks.
This is discussed later in this paper.
1.4.1. Investigating Active Routing Attacks

One unique aspect of the attacks is that Georgia gets nearly all of its Internet access
from two main countries: Russia and Turkey, with some additional connectivity from
Europe. Analysis by Bill Woodcock at Packet Clearing House shows that nearly all of
the major connectivity routes go through Turkey or 0Russia. This provides a high
bandwidth connection for Russian bots, if they are located in Russia, to flood Georgia.
Turk Telecom, the main upstream for Georgia in Turkey, is also a major source of
bots.
Russian ISPs were accused during the fighting of filtering or blocking Georgian
sites, which would have been possible for some routes but not all. In our analysis, we
have found no data that suggests that Russian ISPs performed such filtering [[12]].
Routeviews monitors did see some unexplained BGP announcements via Turk
Telecom but we attribute those to fighting the DDoS traffic or drops due to congestion,
rather than active attempts to disrupt normal Georgian traffic.
Our measurements indicate that approximately 100 BGP updates per day occurred
for Georgian prefixes immediately before the onset of ground fighting with Russia.
After ground fighting began, less than 10 BGP updates per day were seen. The August
cyber attacks began within 24 hours of Russian tanks rolling into Georgia, making the
data hard to decipher conclusively. Any BGP disruptions could be due to fighting on
the ground, DDoS attacks (and congestion leading to drops), or active disruptions by
upstream peers.
1.5. Democratic Voice of Burma
Starting in the summer of 2008, DDoS attacks were launched against the Burmese
dissident site the Democratic Voice of Burma (DVB) and its sister sites. Many of the
attacks were website defacements and the attackers got in through a poorly configured
and poorly secured site. ATLAS monitors recorded some packet flooding to the sites,
as well [[14]].
Most of the attacks were apparent attempts to censor the sites and to thwart
planned 8-8-2008 protests around the world. 20 years before on August 8, 1988, a
significant protest occurred in Burma against the ruling Junta centered on the 8-8-88
date. The number 8 is very significant in Chinese and Burmese society, providing the
protests on 8-8 are a powerful rallying point. The Burmese government is believed to
be behind the attacks, although no such evidence has been provided.
1.6. Russian Elections, 2007
In the lead up to the Russian elections in late 2007, the website for the dissident
politician and well-known chess Grand Master Gary Kasparov and his political party
were both hit with substantial DDoS attacks. Kasparov has been a very vocal
counterpoint to the powers in Moscow, specifically former Russian president Putin’s
administration, for many years. During the attacks, Kasparov’s site was inaccessible,
and so was his political party’s [[13]].
The attack command activity traced back to botnets possibly run by Russian or
pro-Russian hackers. The botnets have been used in the past to strike political targets
among other targets.
1.7. Radio Free Europe/Radio Liberty
In April 2008, Radio Free Europe and Radio Liberty (RFE/RL) websites were hit with
DDoS attacks [[15]]. It is thought that the attacks were in retribution for the reporting
that RFE/RL made to cover the anniversary of the Chernobyl disaster.
The attacks started on April 26 and first targeted the website of RFE/RL’s Belarus
Service and quickly spread to other RFE/RL sites. Within a few hours, eight different
RFE/RL websites serving Belarus, Kosovo, Azerbaijan, Tatar-Bashkir, Radio Farda,
South Slavic, Russian, and Tajik-language listeners were all affected by such attacks.
The botnet behind the attacks was a Russian-language botnet that had been active
in other politically motivated attacks in the recent past.
1.8. Ukraine Anti-NATO Protests
In March 2008, various Ukrainian newspaper sites were hit with DDoS attacks due to
internal political tensions. The C&Cs behind the attacks were located in the Ukraine
[[13]], although it is possible that outsiders or parties operating within the Ukraine used
these botnets.
Also in 2008, the website for ‘5.ua’, a news website for Ukraine, came under
attack with the message “NATO go home” in the HTTP request as part of the flood.
These attacks coincided with street protests against NATO expansion into the Ukraine.
ATLAS monitoring tracked the C&C behind the attacks in this case to the hosts ‘my-
loads.info’ and ‘ultra-shop.biz’, a BlackEnergy botnet controlled located (at the time)
in China that uses multiple names for the same IP address [[14]].
1.9. Kazakhstan Government Criticism, MSK Forums
In early 2009, the forums for the Russian website MSK came under denial of service
attacks. It is believed that these attacks were in retribution for the MSK site posting a
PDF copy of a newspaper that was censored through the Kasakh government by pro-
Moscow forces. The newspaper published an article written by the Kasakh president
that was critical of the Russian government. When no other newspaper would carry the
article, MSK offered to host it online and came under attack shortly thereafter [[16]].
The MSK site forums, in response to the DDoS attacks on site in conjunction with
the Kazakhstan newspaper, hosted a poll on who people thought were responsible for
the DDoS attacks. The poll, dated March 2, 2009, asked, “Who do you think organized
DDoS-attack on forum.msk?” The results speak very significantly at the amount of
distrust in the region:
Kremlin (185)
FSB (121)
Pro-Kremlin youth organizations (68)
MIA (4)
Administration of the Moscow region (3)
Administration Himok (11)
Communist Party (14)
Simple network hooligans (21)
Anti-power (23)
Neo Trotsky Fighters (22)
Other (15)
At this time it is still unclear what group launched the attacks, although ATLAS data
indicates the attacks were lead by the botnets hosted on the sites ‘candy-country.com’,
‘22x2x2x22.com’, and ‘sexiland.ru’. All three of these are identified BlackEnergy-
based botnet controllers.
1.10. Russian Opposition Websites
In late December 2008, a related attack struck the newspaper sites ‘grani.ru’, ‘ikd.ru,’
(which publishes news about demonstrations going on around Russia) and ‘nazbol.ru’
(the website of the banned National Bolshevik Party) [[17]]. All of these attacks are
consistent with the basic premise that the opposition is routinely censored by
DDoS. Data gathered by Arbor Networks indicates that some of the same botnets
behind the MSK attacks (above) participated in these attacks.
1.11. Israel-Gaza/Hamas
During the Israeli-Hamas fighting in Gaza in January of 2009, multiple cyber attacks
were launched both from Israeli hackers and Palestinian (and pro-Palestinian) attackers.
The bulk of the attacks were website defacements, although we did see some DDoS
attacks [[18]]. This is not the first time such cross-border cyber-attacks have occurred.
In fact, the long-standing Israeli-Palestinian conflicts are the source of many such
attacks and the cause for many website defacements on both sides of the conflict.
One of the tools distributed during these attacks was the “Patriot DDoS tool” from
the website “Help Israel Win”. The tool was loaded onto a number of websites and
domains and was routinely shut down by various groups. It had also undergone a
number of iterations to fix bugs and evade any antivirus detection. This is another
example of the voluntary cyber attacks sometimes observed in the wild during
diplomatic conflicts and shooting wars.
1.12. Kyrgyzstan, January 2009 – False Positive?
In mid-January 2009, reports started appearing that the small former Soviet Bloc nation
of Kyrgyzstan was under a cyber attack. The data so far consists mainly of a few
NetFlow logs and some web server logs of a few sites in Kyrgyzstan, but very little
else. The main site reporting this attack, in a blog posting by Secure Works researcher
Don Jackson, blamed the Russian government for the attacks [[19]]. This was followed
up on the IntelFusion blog with some analysis and speculation as to the causes behind
any such attacks [[20]].
In a posting on January 30, the author at IntelFusion made a case that the
Kyrgyzstan government itself launched the attacks [[21]], basing this on some
speculations that are consistent with the events in the region. While many researchers’
attention in the United States was drawn to the threats at the time to close the Manas
airbase (vital to NATO and US efforts in Afghanistan), events within Kyrgyzstan
reveal another story. Instead, IntelFusion’s analysis suggests that it was an effort to
silence critics, since the Kyrgyzstan government is already very pro-Moscow and will
happily comply with any offers that Moscow wields. Indeed, Moscow did openly offer
Kyrgyzstan money if they closed the Manas air base.
ATLAS data was unable to discover independent data to suggest attacks came
through the usual routes such as botnets and coordination via forums [[22]]. ATLAS
data also did not show any Internet backbone flow data that suggests that the attacks
crossed the normal channels.
1.13. Kommersant, 2008
On March 14, 2008, The Kommersant newspaper had complained to police and
prosecutors about a massive hacker attack on its web site, which it suspected was
orchestrated by the pro-Kremlin youth group Nashi. Nashi is one of several youth
groups in Russia that has been involved in street protests and highly organized
activities. They are also suspected in several online attacks including the ones against
Kommersant. At the time, the Kommersant paper had published articles critical of
Nashi and the government and came under fire, possibly in retaliation for this
reporting. ATLAS data tracked several botnet C&C servers issuing commands to their
BlackEnergy-based botnets to launch attacks against the Kommersant servers [[23]].
During the attacks, the Kommersant website was moved to the UK for improved
hosting, although the attacks continued after the relocation.
1.14. Kazakh opposition websites allegedly under DDoS attacks
In February 2009, a Kazakh newspaper website came under attack for publishing
material critical of the government in Astana [[24]]. The newspaper’s site,
‘zonakz.net’, had published articles and recordings of several government officials
purportedly committing crimes and acts of corruption. The site was first shut down in
Kazakhstan and then moved overseas where it came under a DDoS attack.
In a report titled “The Contradictory State of Kazakhstan” that appeared on the site
EurAsia.net, reporter Bruce Pannier wrote about the attacks [[25]]:
Critics claim there is ample evidence of increased scrutiny of media outlets -- whether
traditional or Internet-based.
The owner and editor in chief of the independent weekly "Almaty-Info" is currently on trial
for divulging state secrets in a November 2008 article, and is also being sued for defaming a
businessman.
Also this week, the head of the zonakz.net website complained that Kazakh law enforcement
agencies were blocking access to the website, which is known for having carried material
critical of, and at times potentially damaging to, the government.
After a shutdown of zonakz.net㩟s domestic servers that followed its posting of purported
recordings and transcripts of senior Kazakh officials㩟 phone conversations, the site was
registered abroad only to find access blocked by a new distributor-denial-of-service program
known as DDOS-attack.
Additionally, various political parties have described DDoS attacks against news
outlets in Kazakhstan as a means of silencing political opponents [[26]].
1.15. Iranian Elections, 2009
Beginning in mid-June, 2009, Arbor Networks began to see signs of Internet attack
activity following the disputed presidential elections in Iran [[32]]. Street protests were
organized using online forums and especially the Twitter service, and DDoS attacks
against Iranian media and government sites began almost immediately. Most of the
attacks used simple “page reboot” scripts, which are websites that construct a
repeatedly reloading web page for an attacker that can be used by just browsing to the
website. To maximize their effect, attackers coordinated the timing of their efforts
using Twitter. However, attackers just as quickly suggested the attacks stop due to
bandwidth consumption issues in light of the country’s Internet traffic filtering. It is
unclear if the attacks had any significant impact on the target sites’ availability.
1.16. Coordinated South Korean-US Attacks, July 2009
Beginning on July 4 2009, a series of DDoS attacks began to strike first South Korean
and then both South Korean and US government and commercial websites [[33]]. Sites
targeted included the Korean Assembly, the US and South Korean presidents’ websites,
the US State Department, the public websites for the US stock exchanges NYSE and
NASDAQ, and popular sites in South Korea such as ‘naver.com’. Investigations
revealed a botnet that was apparently built using a variant of the MyDoom worm from
early 2004 together with rudimentary DDoS attacks such as HTTP request floods, UDP
and ICMP floods. The attacks continued from July 4 until July 10, when the infected
PCs were programmed to encrypt files and render themselves unbootable.
The targets, the US and South Korea, together with the timing between a North
Korean missile test launch on July 4 and the 15th anniversary of North Korea’s Kim Il
Sung’s death on July 8 lead some to suggest that North Korea was behind the attacks.
To date, we have not seen any evidence of this. The real motivations for these attacks
remains a mystery, but it is widely considered a political attack.
2. Attackers’ Motivations
In many of the above cases, classic right-wing sentiments are apparently behind the
attacks. In most cases, we appear to see attackers using DDoS attacks to express
support of an official government position, either against external or internal foes. This
is analogous to street protests organized by a political party to stifle opposition through
a show of force. Increasingly, we are seeing DDoS attacks used to silence opposition
sites, such as in the Kommersant attacks, the attacks on MSK, and the recent attacks in
Kazakhstan. A notable exception is the Iranian attacks in June 2009, where anti-Iranian
government protesters apparently organized a series of DDoS attacks to protest the
election results. The July 2009 attacks on government sites in South Korea and the US
may have been a protest, but it is unclear at this time.
In many of these situations, the attacker is able to employ classic guerilla warfare
tactics to grow their size and power through the use of propaganda that appeals to an
ethnic or national base. In these conflicts the attackers first answer the rally call at the
beginning of diplomatic or military hostilities to begin their attacks. They then extend
this force by providing easy to use tools through an extensive network of social forums
and media including blogs, bulletin boards, and specialized information sites (often
dubbed "inform" sites by the Russian hacker underground). Materials posted and re-
posted here encourage new recruits to seek retribution against their enemies and join
the fight. What starts as a small, core group is can grow into a massive force.
Propaganda effects can be so strong, and long lasting, that Estonia still watches for
renewed attacks every year on Victory Day. They have seen some attacks but nothing
that rises to the level of the 2007 attacks.
By using cheaply and widely available technology, the enemy can leverage IP
protocols, botnets, and applications as a force multiplier. That is to say that by using
such tools attackers have a reach and power significantly beyond their normal capacity.
The techniques to launch these attacks are commonly discussed; fortunately any
advance in the sophistication of these techniques is much slower. However the
attackers are able to codify their methods into easy to use tools that can be shared
freely. There is an increasing emphasis on the ease of use for these tools by outsiders or
non-technical parties. An example is the appearance of websites that use dynamic

HTML methods to launch HTTP floods simply by loading a specific website. These
tools were popular in the recent DDoS attacks on the Iranian government following a
disputed national election, commonly using the website ‘pagereboot.com’.
3. Attackers’ Aims and Goals
Historically, these DDoS attacks have been aimed to cause the victim some punitive
damage or register their dissent with the victim’s actions. These are the apparent
motivations in the attacks from Chinese hackers in retaliation for the embassy bombing
in the late 1990s, and the 2007 Estonia attacks, the 2008 Georgia attacks, and the 2009
attacks on Iranian websites. We have seen changes with recent attack activity. Lately,
the apparent goal of the attacks is to censor the opposition, either a dissident populace
within the country, or dissidents outside the country, or an adversary elsewhere in the
world. These are the kinds of attacks we see in the Russian elections of 2007 and
subsequent attacks.
The Internet has become a major communication tool for news media,
governments, political parties, the opposition and dissidents. Striking at their voice,
their printing press, and their Internet channels makes perfect sense. This is apparently
the main motivation of the attacks against the Democratic Voice of Burma, where a
coordinated series of website hacks and defacements, as well as some DDoS attacks,
were used to disrupt global protests against the ruling military in Myanmar.
The cheap and easy availability of the tools and weapons - botnet armies, hacker
groups, and the like - have caused governments around the world to eye this approach
as a means of silencing enemies. Even when there is no direct tie to the government,
such actions can benefit the ruling party’s aims. However, in every case we have been
unable to conclusively say that the government has been behind the attacks. If
governments use such tactics and tools in modern information warfare, then these
attacks, by using independently operated botnets, make an excellent attack tool with
plausible deniability for the attack director.
4. Attribution
Many have accused government actors or sponsored actors of carrying out these sorts
of DDoS attacks. It is important to note that we cannot attribute any of these attacks to
a specific group or agency with our data. We simply do not have the evidence to
confirm it. All analysis of the data we have suggests non-state actors, however. This
comes from observing the attack through three major means: direct data observations,
community discussions encouraging and organizing the attacks, and analyzing the
botnets and tools used to conduct the attacks.
In a LiveJournal account that we spotted we read representative during the denial
of service attacks on Estonia in 2007 [[27]]. The post contains a simple DOS batch
script that lists Estonian servers and IP addresses to be ping flooded and enters an
infinite loop. The messages around the posting, and in similar forum postings, describe
the Estonians as “fascists”, “amateurs”, and saying that they must be attacked.
Based on flow data from one of the attacks during the Estonian incident, we
mapped where the traffic origins to geographic coordinates. The result quite clearly
shows how widely distributed the attacks were sourced, namely from all over the
world. In this case this particular attack was from a botnet. We do not think that this
attack used source spoofing as all of the IP addresses in question mapped back to
allocated netblocks and not unallocated IP address space, as is commonly seen when
the attacks used spoofed or forged source IP addresses.
Some of the attacks were from far more discrete sources and likely came from the
ping flood scripts that were in circulation. These were run by far fewer people and
therefore had a smaller base of hosts to come from. We identified these attacks by their
traffic type, ICMP echo request, and by the networks the traffic sources aggregated to,
network allocations in Europe and Russia.
During the investigations into who launched the attacks, a 20-year-old Estonian
student was charged and fined for his part in the attacks [[28]]. His fine was very small,
only about $1650. Based on our data showing botnets, ping flood scripts, and the
attackers’ discussion, we conclude that it is unlikely that Dmitri Galushkevich is the
only person responsible for the attacks, however.
Attribution continues to a significant challenge in this problem space when
retaliatory measures are considered. In the July 2009 attacks on South Korean and US
websites, the South Korean intelligence services stated through the press that they
suspected North Korean hackers were behind the attacks. This was picked up and used
as a call for retaliation on North Korea by a US lawmaker a few days later. Clearly,
these kinds of attacks can spiral into significant diplomatic incidents if great care is not
taken.
4.1. Role of Russian Youth Groups
An examination of recent attacks shows that in many cases there are political
skirmishes with Russia at the core of the attacks. In these scenarios, one commonly
fingered segment of the Russian hard-line community is political youth groups. These
organizations are partially state-sponsored and used to hold pro-Kremlin rallies, but
have also been accused in various physical attacks over the years. As noted earlier in
this paper, they have been accused of the Kommersant attacks, among others. The
Russian youth group Nashi claimed responsibility for the Estonian attacks of May 2007
in a news report from mid-2007 [[29]].
Claims about who was behind the Estonia attacks in 2007 were renewed during a
2009 videoconference between Moscow and Washington, and was described in a news
report [[30]]. The participants talked about the methods and technologies of
information warfare in the 21st century, based on examples of the “Inform Campaign”
model that accompanied the military and economic conflicts in recent years (the five-
day war in Georgia in August 2008, Israeli military operation in Gaza in early 2009, the
gas delivery conflict between Ukraine and Russia, etc.). “Inform campaigns” are
routinely used to coordinate such attacks and are widely thought to be government
assisted if not outright sponsored.
Sergei Markov, a State Duma Deputy from the pro-Kremlin Unified Russia, claimed in a
March 3, 2009, discussion that his assistant was responsible for the attacks. Said Marvov,
“They did not know what to do next. There were feasts, to whom they could not reach. They
call to me and say: Sergey, what to do now? Here, we have disabled Estonian sites. I do not
know what to do! I say: So what? Let's let this information that is learned.” Markov reportedly
said ominously, "and, incidentally, such things will happen more and more.” Nashi, the
Russian youth group, renewed their claim of a role in the attacks as well.
”In this way, the boys expressed their protest against the policy of the state of fascism carried
out by the leadership of the Republic of Estonia”, - quoted Commissioner movement
Webplanet.ru.
4.2. Hainan Island incident
The Chinese hacker group “Honker Union” took credit for the 2001 hacking incidents
in relation to the Hainan Island incident, including the DDoS attacks and the probes on
US government computers. This claim is widely believed to be accurate [[4]]. Honker
Union is now merged with another Chinese hacking group. Such groups appear to
operate openly in China and can sometimes organize such political attacks.
4.3. Botnets behind Georgia-Russia Cyber War
Many of the botnets we listed above, and more, actively participated in attacks against
Georgian websites. We recorded well known as well as new BlackEnergy-based
botnets striking Georgian targets, most launching generic flood attacks. We identified
only a few botnets launching attacks into Russia.
One of the sites set up to coordinate cyber-attacks on Georgia as well as to share
ongoing information about the war was the site ‘OSInform.RU’. The website contained
imagery of death and skulls, and also claims of genocide, material seen consistently in
sites set up by Russian hackers detailing attacks on Georgian sites. Multiple blogs
begin sharing a simple ping flood scripts targeted Georgian sites, a very similar scripts
to the ones seen in Estonia.
A “Stop Georgia” site was set up to coordinate cyber attacks on Georgian web
properties. Self appointed representatives of the Russian hacker underground claimed
to be behind the site, and it was hosted in multiple locations (via mirroring). The
translated comments on the site were:
Our response to aggression by Georgia
We - the representatives of Russian hacker underground 0 will not tolerate

provocation by the Georgian in all its manifestations. We want to live in a free
world and exist free from aggression and lies space. We do not need the guidance
from the authorities or others, but act according to their convictions based on
patriotism, conscience and belief in the virtue of justice. You can call us
criminals and cyber-terrorists, continuing with war and killing people. But we
will fight and unacceptable aggression against Russia in cyberspace.
We demand the cessation of attacks on information and government resources on

RUNET, as well as appeal to all media and journalists with a request to cover
events objectively. Until the situation has changed, we will impede the
dissemination of false information by the Georgian government and information
resources. We did not launch an information war, we are not responsible for its
consequences.
We call for the assistance of all who care about the lies of Georgian political
sites, everyone who is able to inhibit the spread of false information.
StopGeorgia.ru
P.S. There is one formal mirror project - www.stopgeorgia.info. All other

resources have nothing to do with the movement StopGeorgia.ru.
The “Stop Georgia” site also contains a list of sites belonging to Georgia government
agencies or Georgian properties abroad. The exhaustive list provides victim IP
addresses for targeting and shows their status.
Russian attackers had significant coordination to their activities that was quickly
set up, many within a day of the ground offensive beginning. We are not clear on the
timelines of the buildup of border tensions or any propaganda campaigns by Russia
against Georgia, although a significant lead up to the shooting war could have allowed
attackers to establish their operations in time for the ground hostilities.
5. Official Responses Since Estonia
The spring 2007 events in Estonia have served as a clear wake up call to governments
around the world about the power of cyber attacks and the damage they can inflict. The
events in the summer of 2008 against Georgia were a forceful reminder of the attacks
and added great urgency to this analysis. Many governments are reviewing their own
vulnerability to DDoS attacks or more common infiltrations. A small handful of nations
are investigating active cyber attack programs of their own.
5.1. Defensive Responsibilities
Especially since May 2007, but even more after the 2008 Georgia attacks, governments
and groups around the world are worried about being a victim of a cyber attack.
NATO, the EU, and other groups have been investigating their role in responding and
their responsibilities and obligations. To date neither the EU nor NATO has articulated
clear strategies for countering such attacks on member states.
The IMPACT alliance (http://www.impact-alliance.org/) has been founded in
Malaysia to combat cyber terrorism and has been working to become a UN of cyber
security, in part with the help of the ITU.
5.2. Role of Attribution in Response
Attribution is a key aspect for any large-scale response including retribution attacks or
seeking redress via the international community, such as in the UN or via diplomatic
channels. These kinds of attacks give a nation-state clear plausible deniability if they
are actively sponsored, and an even bolder claim if these are simply run out of the
civilian populace but tolerated or even tacitly controlled.
Some have claimed that the use of subtle language cues is commonly employed by
the Chinese to direct such attacks. Phrases that seem innocent can have a sweeping
impact on how the populace responds, either in street protests or in online attacks. If
this is the case then we should expect that these kinds of attacks would continue and
become a tool for managing opposition or foes in the 21st century. Their impact -
bandwidth, durations, victims - is likely to grow and their frequency, scale, and the
number of origins is likely to grow as well, as we have seen in the past several years.
6. Recommendations
Recent history has shown that packet flooding attacks are increasingly a favorite
weapon of politically motivated attackers regardless of their geographic region. These
attacks threaten communication mechanisms, the integrity of elections, and the freedom
of an independent press, the activities of dissident groups and politicians, and may, in
the future, grow in sophistication and disrupt normal daily life. In this time we have
seen investigations and defense measures spawned from independent parties, the
commercial sector, and the government sector through mostly ad-hoc means. While
this has been marginally effective so far, this has quickly become an untenable
situation.
A number of recommendations follow based on the author’s experience in a
number of the conflicts described above.
6.1. Broad Defensive Contributions Must be Possible
If we are to successfully defend national infrastructure against the sorts of attacks that
affected Estonia and Georgia then we must be open to all forms of assistance. In both
cases the public were firmly on the side of the victim (Estonia, Georgia), a sentiment
that must be harnessed more effectively in the future. This must be turned into
Schwerpunkt - a unity of purpose and goals - which will make us effective in our
mission of defending the Internet.
Commercial tools from various vendors, including the author's employer, exist to
detect and filter DDoS attack traffic and have been deployed to help thwart some of the
attacks reviewed above. The technology in these tools is commonly available and the
only barrier to their deployment is budget. However, as a total solution to the political
DDoS problem this is insufficient from a cost or management perspective. We must
think about how to utilize new methods to defend critical and civilian infrastructure as
well as government infrastructure.
The enemy, attackers, uses public sentiment on his side to grow an organic legion
of supporters to aid in their cause. Their aim is more amorphous than the defenders'
role but the principle applies: by utilizing propaganda campaigns and nationalist and
ethnic sentiment, he grows his army of volunteers. This is exactly analogous to the
enemy in guerilla warfare.
Defenders do not use organic support for their mission of stopping these attacks,
however. Outside support has been used to some extend in the recent past, with Tulip
Networks in Atlanta, Georgia, in the United States providing bandwidth and
connectivity for some of the Georgian infrastructure under attack. This was made
possible through a direct, personal friendship that enabled this help. This kind of
assistance is rare and no formal agreements are in place, leaving victims at risk.
For the victims, successes in defending an online presence usually come when a
group or an individual acts on his or her own with the best interests in mind. Many
more individuals or groups who could help are usually blocked from providing
assistance. More outsiders are willing to help in these cases through meaningful ways,
and we must enable them to provide aid if we are to defend these networks and this
infrastructure. One challenge that will have to be addressed is to discover which offers
are credible or worthy. However, a network of professionals to defend against these
sorts of attacks exists in the commercial Internet service provide realm.
Governments must be open to assistance from the private, commercial sector for
dedicated DDoS-resilient hosting for public facing Internet properties. At this time the
targets of these attacks mainly consist of information-only sites, but in the future will
surely include key infrastructure equipment such as VoIP exchange points, DNS
servers, and email systems which, if targeted, could impact the ability of a government
to communicate internally. Governments and other likely political targets such as
newspapers must identify how they can migrate their infrastructure to a third-party’s
systems to ensure continuity.
Furthermore, governments and targets must be trained and willing to accept a rapid
deployment of commercial tools to defend against these kinds of attacks. All members
of the government’s information technology staff should be able to receive an offer of
help and determine its credibility, and route that offer to the appropriate internal party
for follow up. We have seen this work in limited cases in the past but too often we find
that government victims in these attacks do not know how to accept an offer of
assistance in a timely fashion.
6.2. Improved Efficiency in the Decision Making Process
A review of the OODA loop, or the Boyd cycle, provides ample areas to review and
seek improvement in our current posture [[31]]. The cycle is built of four core steps
that provide feedback to each other: observe, orient, device, act. The faster and more
accurately one side can complete the loop - and begin the cycle again - the bigger an
advantage he has.
Our observation points are currently piecemeal and hampered by competing
business interests. This is nothing new, but it means we have a poor foundation on
which to base our decisions. Because we lack a complete overview of Internet activity
about the origins of attacks and how we may stop them, we often waste valuable time
defending against attacks when we could stop them at their root. Information
collection, sharing, and recall are woefully ignored and falling behind.
As a community of defenders we are usually able to orient at the broader goal -
defend a specific country's assets (e.g. Estonia), identify the attackers behind it - but
our more specific tactics to achieve that goal are unfocused and lacking. We fail to
communicate what we need, what we find, and what the next steps are.
Our decision making process is often mired in consensus building and dogged by
second-guessing. We are ineffective in many cases because we fail to make decisions
for fear of making the wrong one. Committees with the wrong stakeholders and people
who have no value to the process hijack and derail the process.
Finally, our actions are bound by laws and jurisdictions but also by seeking the
permission of too many parties. In short, we move too slowly, too blindly, and too
ineffectively, if we move at all. We are not consistently effective.
Moving forward, governments and coordination centers must be given the
authority to act without requiring a consensus of all parties but rather act quickly in the
best interests of the group. This should be treated as an authority akin to a military
command authority and should coordinate public-sector, private-sector, and military
efforts at combating attacks. Careful balance must be taken to work with carriers, for
example, to avoid disruptions to the infrastructure, a key facet to ensuring the carriers
will accept outside leadership in such events.
7. Conclusions
DDoS attacks provide a simple, easily available mechanism to disrupt the Internet
presence of a group or a small nation. Previously, they have been confined to
retaliatory attacks seeking punitive damage to the victim, but in recent years the role of
the Internet in publishing newspapers or organizing dissident efforts has grown. The
growing importance of the Internet to potential victims has not escaped cyberwar
practitioners. DDoS attacks will continue as a tool of censorship as long as the Internet
remains a communications medium.
Cyber-warfare takes on different forms in different areas of the world. Political
targets and motivations in DDoS attacks are most popular in Russia and the region, less
so in China, Asia and the Middle East. China favors more surgical, infiltration events
for serious cyber warfare. We have seen an explosion of DDoS tools from Chinese
hackers, although most of their targets are commercial sites located in China, but many
are in Korea or Japan. These sites are the targets of bullying or extortion attacks that do
not yet rise to the level of political warfare. Burma benefits from website defacements
and destruction. Israel and Palestine often use website defacements to challenge each
other. At this time we expect to see DDoS attacks continue to be a political weapon in
the Russian power sphere, particularly for former Soviet bloc nations.
These attacks will continue to provide the nation-state benefits from their actions
as well as plausible deniability should they actively engage in such actions. Because of
this we expect their frequency to grow in the Russian region, together with their
sophistication as victims begin to develop improved defenses. Furthermore we
anticipate that other nations may begin using DDoS attacks as a simple, blunt force
political weapon to silence critics or opponents.
Much of the theory of cyber-warfare remains to be written, but may borrow from
other warfare theories. Specifically theories on guerilla and asymmetric warfare need to
be reviewed to understand the enemy’s tools and tactics, as well as to understand
responses. While governments and private industry control the communication’s fabric,
they have yet been unable to muster a unified, consistent defense. Instead, defenses
have largely been ad-hoc and at the mercy of generous outsiders. Responses must be
cohesive if not unified in order to be consistent, an approach that would be well
informed with an understanding of defense tactics learned from studying theories of
cyber-warfare.
References
[1] Mirkovic, J. and Reiher, P., A taxonomy of DDoS attack and DDoS defense mechanisms, in ACM
SIGCOMM Computer Communication Review, 2004.
[2] Arbor Networks Website, http://www.arbornetworks.com/en/products.html.
[3] ACTIVISM, HACKTIVISM, AND CYBERTERRORISM: THE INTERNET AS A TOOL FOR
INFLUENCING, Denning, D.E., in Networks and netwars: The future of terror, crime, and
militancy, 2001.
[4] Cyber Protests: The Threat to the U.S. Information Infrastructure, National Infrastructure Protection
Center, 2001. Available online at http://www.au.af.mil/au/awc/awcgate/nipc/cyberprotests.pdf.
[5] US State Department Website. Available online at http://www.state.gov/r/pa/ei/bgn/5377.htm.

[6] Estonian DDoS Attacks - A summary to date, by Jose Nazario, on Security To The Core weblog,
May 17, 2007. Available online at http://asert.arbornetworks.com/2007/05/estonian-ddos-attacks-a-
summary-to-date/.
[7] CNN Attack Summary, by Jose Nazario, on Security To The Core weblog, April 21, 2008.
Avaiable online at http://asert.arbornetworks.com/2008/04/cnn-attack-summary/.
[8] Cyber Attacks Against Georgia: Legal Lessons Identified, by Eneken Tikk, Kadri Kaska, Kristel
Rünnimeri, Mari Kert, Anna-Maria Talihärm, and Liis Vihul, 2008.
[9] Georgia On My Mind – Political DDoS, by Jose Nazario, on Security To The Core weblog, July 20,
2008. Available online at http://asert.arbornetworks.com/2008/07/georgia-on-my-mind-political-
ddos/.
[10] The Website for the President of Georgia Under Attack - Politically Motivated? by Steven Adair, in
Shadowserver Foundation Calendar, July 20, 2008. Available online at
http://www.shadowserver.org/wiki/pmwiki.php/Calendar/20080720.
[11] Estonia hosts Georgian Web sites to halt hackers, on FoxNews.com, August 26, 2008. Available
online at http://www.foxnews.com/wires/2008Aug26/0,4670,EstoniaGeorgiaHaltingHackers,00.
html.
[12] An In-Depth Look at the Georgia-Russia Cyber Conflict of 2008, Jose Nazario and Andre M.
DiMino, at the Botnet Task Force meeting, October 2008.
[13] Political DDoS? Ukraine, Kasparov, by Jose Nazario, on Security To The Core weblog, December
13, 2007. Available online at http://asert.arbornetworks.com/2007/12/political-ddos-ukraine-
kasparov/.
[14] Political DDoS: Estonia and Beyond, Jose Nazario, in a presentation give at Usenix Security, 2008.
Available online at http://www.usenix.org/events/sec08/tech/slides/nazario-slides.pdf.
[15] Radio Free Europe DDoS, by Jose Nazario, on Security To The Core weblog, April 29, 2008.
Available online at http://asert.arbornetworks.com/2008/04/radio-free-europe-ddos/.
[16] MSK Forum, February 28, 2009, available online at http://forum.msk.ru/notice/.
[17] Russian Opposition Websites Shut Down By Attacks, December 25, 2008, on The Other Russia.
Available online at http://www.theotherrussia.org/2008/12/25/russian-opposition-websites-shut-
down-by-attacks/.
[18] The Effects of War: Gaza and Israel, by Jose Nazario, on Security To The Core weblog, January 5,
2009. Available online at http://asert.arbornetworks.com/2009/01/the-effects-of-war-gaza-and-
israel/.
[19] Kyrgyzstan Under DDoS Attack From Russia, by Don Jackson, on SecureWorks Research Blog,
January 28, 2009. Available online at http://www.secureworks.com/research/blog/in-
dex.php/2009/01/28/kyrgyzstan-under-ddos-attack-from-russia/.
[20] The Kyrgyzstan DDoS Attacks of January, 2009: Assessment and Analysis, Jeff Carr, jart Armin
and Greg Walton, on IntelFusion blog. Available online at http://intelfusion.net/wordpress/?p=516.
[21] Why I believe that the Kyrgyzstan Government hired Russian hackers to launch a DDOS attack
against itself, by Jeff Carr, on IntelFusion blog. Available online at
http://intelfusion.net/wordpress/?p=520.
[22] Kyrgyzstan DDoS Attacks, by Jose Nazario, on Security To The Core weblog, February 2, 2009.
Available online at http://asert.arbornetworks.com/2009/02/kyrgyzstan-ddos-attacks/.
[23] Russian DDoS Attacks: Kommersant, by Jose Nazario, on Security To The Core weblog, March
19, 2008. Available online at http://asert.arbornetworks.com/2008/03/russian-ddos-attacks-
kommersant/.
[24] Quick Notes on Cyber Warfare News, by Jose Nazario, on Security To The Core weblog, February
19, 2009. Available online at http://asert.arbornetworks.com/2009/02/quick-notes-on-cyber-
warfare-news/.
[25] THE CONTRADICTORY STATE OF KAZAKHSTAN, by Bruce Pannier, in EURASIA
INSIGHT, March 6, 2009. Available online at
http://www.eurasianet.org/departments/insightb/articles/pp030609d.shtml.
[26] Kazakhstan: Five political parties report about the information terrorists to the public prosecution
office, February 25, 2009, on the website Ferghana.ru. Avalable online at
http://enews.ferghana.ru/news.php?id=1024.
[27] “Load quickly on chuhonofilam”, in a posting on a LiveJournal blog by w8kl8dlaka. Available
online at http://w8lk8dlaka.livejournal.com/52383.html.
[28] Student fined for attack against Estonian Web site, Jeremy Kirk, InfoWorld, January 24, 2008.
[29] Nashi, Russia's new militant nationalist movement, Rediff India Abroad, May 21, 2007. Available
online at http://www.rediff.com/news/2007/may/21nashi.htm.
[30] Behind the Estonia Cyberattacks, Radio Free Europe/Radio Liberty, March 6, 2009.
[31] Osinga, Frans. Science, Strategy and War: The Strategic Theory of John Boyd. Abingdon, UK:
Routledge, 2007.
[32] Iran DDoS Activity: Chatter, Tools and Traffic Rates, by Jose Nazario, on the Security to the Core
weblog, June 19, 2009. Available online at http://asert.arbornetworks.com/2009/06/iran-ddos-
activity-chatter-tools-and-traffic-rates/.
[33] Korean/U.S. DDoS Attacks – Perplexing, Disruptive, and Destructive, by Steven Adair, on the
Shadow Server Foundation Calendar blog on July 10, 2009. Available online at
http://www.shadowserver.org/wiki/pmwiki.php/Calendar/20090710.
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-182
A Brief Examination of Media Coverage of

Cyberattacks (2007 - Present)
Cyrus FARIVAR
Freelance Technology Journalist (NPR, PRI, CBC, The Economist)
Abstract. As cyberattacks become more frequent, they draw new attention in the
media. Indeed, there has been a significant spike in journalistic coverage of
cyberattacks and cybersecurity in the last year alone, making this particularly
relevant now. The aim of this paper is to provide an overview of coverage and
make suggestions for future journalists and policymakers to work better together
to better understand this new threat.
Keywords. Media studies, media, journalism, cybersecurity, cyberattacks
Introduction
In the last decade, there have been countless cyberattacks against various political,
military and economic targets in the United States, Europe and elsewhere. Some have
been made public, and others remain classified. Many of these cyberattacks have been
against various American military targets and some have have overlapped into cyber-
espionage territory. (For the purposes of this paper, I will focus only on political-
oriented direct cyberattacks, not cyber-espionage.)
Since 2007, the world has seen three major politically-oriented cyberattacks
(denial of service attacks) against three former Soviet Union countries, Kyrgyzstan
(January 2009), Georgia (August 2008), and Estonia (April-May 2007). All three likely
originated from within Russia, and may have implicitly involved the Kremlin, despite
official denials. As such, this increase in cyberattacks has resulted in a corresponding
increase in the amount of coverage this issue receives in the English-language print
media. In the case of the 2007 attacks against Estonia and the 2008 attacks against
Georgia, both made the front page of The New York Times. However, while there has
been more attention paid to this issue, some of it has been misleading at best and false
at worst. Therefore, it is in the interests of the cybersecurity community and the media
that cover them to better understand how the media has treated cyberattacks, and to
improve public understanding of this phenomenon.
C. Farivar / A Brief Examination of Media Coverage of Cyberattacks (2007 – Present) 183
1. Not all Cyberattacks Are Created Equal (Kyrgyzstan)
On January 28, 2009, The Wall Street Journal ran this headline: “Kyrgyzstan Knocked
Offline.”[1] However, the six-paragraph article, which relied on two sources, only one
of which was named, described how a denial-of-service attack hit the country's two
main ISPs, accounting for nearly 80 percent of the country's bandwidth. While such a
tactic would seem like major news, it was treated as a minor, largely unimportant story.
The Journal relegated it to page A10 of the newspaper, indicating that the news was
only moderately important. The attack was also covered by a few industry publications,
including Computerworld and The Register. The New York Times, ignored the story in
print and only wrote about the event on its blog, The Lede.[2]
This lack of attention shows that when a minor, obscure country gets hit, it's
difficult to develop much interest in such a story – particularly when it's a country that
doesn't have an active online presence, nor that is accompanied by any kind of
corresponding real-world action, nor is it an active member of a multi-national
organization like the European Union or NATO. This is not to say that the attack
against Kyrgyzstan should not have warranted more coverage. If any North American,
E.U., or East Asian nation suddenly had 80 percent of its online capacity knocked out,
it likely would have made international headlines, as it did in late 2008 when an
undersea cable near Egypt was cut by accident, and not as a result of a cyberattack.[3]
This is an unfortunate example of a double-standard in the media should be rectified
the next time something like this happens.
2. When a Cyberattack Accompanies Real-World Events, People Take Notice

(Georgia)
In August 2008, when Georgia suffered a cyberattack that accompanied its invasion by
Russia, the world sat up and took notice. The Wall Street Journal reported: “Georgia
States Hit By Cyberattack,” while The New York Times noted: “Before the Gunfire,
Cyberattacks.”[4] Most media outlets sat up and took notice that a cyberattack element
corresponded with actual physical attacks. Even though these attacks again took the
form of “hacktivism,” and denial-of-service attacks, these media outlets tended to
analyze the online component in more straightforward and plain terms. The Times
noted that the attacks simply “overload and effectively shut down Georgia servers.”
As the second major cyberattack in recent memory, the Georgia attack was notable
as the cyberattack was squarely set in the context of the events on the ground. Perhaps
one of the reasons why the attack against Kyrgyzstan never captivated the attention of
reporters and editors in the same way was because there was no clear narrative of why
it happened – competing theories about obscure political disputes in far-off countries
perhaps don't work. In Georgia, like in Estonia before it, there was a clear example of a
former occupying power asserting its dominance, like a bully beating up a little kid.
This attack was also notable as it was the first (and possibly only) cyberattack where a
journalist became an active participant in the war – albeit in a very minor way. Evgeny
Morozov, a Belorussian journalist now living the United States, in his Slate piece
“How I became a soldier in the Georgia-Russia cyberwar,” showed how easy it was for
an average Russian-speaking Internet used to quickly acquire the tools necessary to
throw an “e-Molotov Cocktail.”[5] Morozov was likely the first journalist who quickly
understood how such an attack could emerge so quickly. In essence, nationalist fervor
184 C. Farivar / A Brief Examination of Media Coverage of Cyberattacks (2007 – Present)
plus an Internet connection could rapidly constitute a “cyberwar.” He concluded his

piece this way, noting:
In less than an hour, I had become an Internet soldier. I didn't receive any calls from
Kremlin operatives; nor did I have to buy a Web server or modify my computer in any
significant way. If what I was doing was cyberwarfare, I have some concerns about the
number of child soldiers who may just find it too fun and accessible to resist.
My experiment also might shed some light on why the recent cyberwar has been so
hard to pin down and why no group in particular has claimed responsibility. Paranoid
that the Kremlin's hand is everywhere, we risk underestimating the great patriotic rage
of many ordinary Russians, who, having been fed too much government propaganda in
the last few days, are convinced that they need to crash Georgian Web sites. Many
Russians undoubtedly went online to learn how to make mischief, as I did. Within an
hour, they, too, could become cyberwarriors.
3. The First Cyberwar (Estonia)
2007 was the first time that The New York Times ever used the word “cyberwar.” In its
May 29, 2007 article, “Digital Fears Emerge After Data Siege in Estonia,” the
American newspaper of record followed the tone that had already been set for much of
the worldwide English-language media coverage of the event.[6] The BBC, which was
one of the first major news outlets to publish, declared on May 17, 2007: “Estonia hit
by 'Moscow cyber war.'”[7] On the same day, the British newspaper The Guardian
wrote: “Russia accused to unleashing cyberwar to disable Estonia.”[8]
While the BBC had used the term before, this was the first time that it had been
used to describe real-life state-to-state attacks. In addition to declaring the events a
“war,” there was a great deal of description about how “E-stonia” essentially
functioned off of its Internet applications. While it is true that Estonia has a high level
of connectivity, Internet banking, online voting and all the rest, the tone of many
articles illustrated a scene of near-meltdown and destruction. The Times reported that
the attacks “came close to shutting down the country's digital infrastructure.” The
Washington Post wrote that the attacks “disrupted government e-mail and led financial
institutions to shut down online banking.” Jaak Aaviksoo, Estonia's defense minister,
told Wired that Estonia's national security was threatened. However, the attacks, while
annoying, did not do any permanent damage, nor was the society in immediate peril.[9]
While there was little technical difference between the attacks against Estonia and
Georgia, the first political “cyberwar,” Estonia's technological landscape made the
rhetoric used that much more dramatic.
4. Lay Off the Hyperbole – It's the Worst Thing Ever
If there is anything to be learned from the first “cyberwar,” or the first “Web War
One,” (as Wired called it) is that hyperbole is a great weapon that can be used
effectively to draw the attention of the world. I'll admit that I myself fell for it – my
Slate piece in the aftermath of the attacks on Estonia was dubbed by my editors as
“Cyberwar I.”[10] In retrospect, the term “cyberattack” would have been more
descriptive, as a war implies a congruous, more or less armed conflict between two
clear entities. In this case, the metaphor of “war” is not very accurate, as it was not
possible for Estonia (or any other cyberattacked country) to retaliate even if it wanted
to. In a cyberattack, the only strategy is defense – there is no way to counter-attack, or
to take out the online firing turret. Furthermore, it's impossible to have a war against an
enemy even more faceless and intangible than international terrorist organizations. If
ordinary, un-technically sophisticated people like Morozov can become
“cyberwarriors” within an hour, does that mean, then, that they are protected under the
Geneva Conventions? Using the language of war quickly breaks down.
In addition to using the term “cyberwar,” everyone has been easily seduced by the
armageddon-style rhetoric that Estonian government officials and associated figures
have used to describe what had happened. Ene Ergma, the speaker of the Estonian
parliament, in an interview with Wired magazine, compared the cyberattacks to a
nuclear explosion, calling them “the same thing.”[11] Linnar Viik, the Estonian
Internet guru, told The Washington Post: “These attacks were an attempt to take one
country back to the cave, back to the Stone Age.”[12] Not only are these statements
ludicrous on their face, but they're blatantly untrue. If the Kremlin or the Russian
“hacktivists” had wanted to pummel Estonia, then the attacks wouldn't have ceased two
weeks after they had begun. The attacks clearly were meant as a message, not as a war.
With all due respect, it was wrong of Ergma and Viik to make such hyperbolic
statements, and it was equally wrong of anglophone journalists to lap it up as easily as
they did. Journalists have a responsibility to not take such ridiculous statements at face
value, particularly ones who have a history of reporting on technology.
Journalists and Estonians alike would do well do remember the example set by
President Bill Clinton in February 2000. This was just after major American tech
companies including Yahoo, Buy.com and CNN were hit with denial-of-service
attacks. In a press conference, the president was asked if this attack was the “electronic
Pearl Harbor.” Clinton replied: “Well, I hope not. (Laughter.) I think it was an alarm. I
don't think it was Pearl Harbor. We lost our Pacific fleet at Pearl Harbor – I don't think
the analogous loss was that great.”[13]
5. Cyberattacks and Civilians
As a technology journalist, or as a cybersecurity professional, it's easy to have tunnel

vision. It's easy to see botnets on every network and miscreants in every Internet forum.
This is not to say that these threats are not real. Rather, it is important to step back from
our bandwidth-fueled lifestyle and begin to examine how cyberattacks do or don't
affect people in the real world. It is a luxury to have high levels of Internet services,
and it is equally a luxury to be able to worry about whether or not these sites are
affected by online “warfare.”
While trying to report on the cyberattacks against Georgia in August 2008, I was
embarrassed when calling the Georgian Ministry of Foreign Affairs in Tbilisi, and a
spokesperson rebuked me for wanting to know about cyberattacks, when in fact the
Ministry was far more concerned with protecting territorial integrity and Georgian
citizens, rather than where the ministry's web site was going to be hosted. While it may
be of great concern and worry to many cybersecurity professionals who have warned
for years of coming cyberattacks – these types of attacks, at least in their current form,
take a back seat to actual, physical warfare. After all, it is worth repeating that no one
has died as a result from a cyber attack. Further, while the Estonian Internet security
community was going into overdrive during the cyberattacks of 2007, the Estonian
public did not seem to be touched by the attacks. In a survey by the Estonian
newspaper Postimees, nearly half (over 49 percent) of the 1,243 Estonian surveyed said
that they were not affected by cyberattacks.
6. Difficulty of Catching the Cyberattackers
If there's one point that should be made to journalists and policymakers alike, it's that
after nearly a decade of major denial of service attacks, that there is neither a perfect
way to secure against them, nor is there a good way to track the perpetrators. After the
attacks against CNN back in 2000, Richard Power, an official of the Computer Security
Institute, told the news network at the time that such attacks “will be one of the most
difficult things to address.”[14] Indeed, it seems that while the attacks may have gotten
more sophisticated and larger, that the basic procedure and execution of such an attack
has not changed hardly at all since an attack that unleashed an estimated 800 megabits
per second of data on web servers. Estonia was only able to defend against the attack
by severing, temporarily, its international data connection to the outside world. Smaller
countries with a limited number of international pipes can employ this tactic, whereas a
much larger online presence like the United States, are unable to.
Further, it should be underscored that it's very difficult to catch anyone who
engages in a cyberattack. Even the attacks against Estonia, which were publicized and
had a high-level of international involvement, have only resulted in the arrest and
successful prosecution of one Estonian citizen, Dmitri Galushkevich.
The 19-year-old quickly confessed to attacking government computer networks,
which is punishable – according to the Estonian Penal Code Section 206, subsection 2
– up to three years in prison.[15] But Galushkevich said that he acted alone, based on
instructions that he read online, which were probably not unlike the ones that Morozov
discovered. He didn’t have any knowledge about who the masterminds or perpetrators
in other countries might be.
It is important to remember that in the immediate months after the 2007
cyberattacks, the Estonian government attempted to request further information from
Russian authorities. Officials had a list of IP addresses that appeared to originate from
within Russia, and needed the help of their neighbor to conduct further investigations,
and perhaps find new suspects. But the Russian Embassy in Tallinn and the Kremlin
gave their Estonian counterparts the run-around, arguing that technicalities of the treaty
between the two countries prevented Russia from providing this information. Further,
the Russian constitution forbids the extradition of its own citizens, so there was no way
for Estonian authorities to question or even depose any Russians. Partly because of the
evidence that he’s seen, and Moscow’s reluctance to be cooperative leads made
Estonian Chief Prosecutor Margus Kurm say that he is confident that the leaders of the
attacks are in Russia, despite saying: “We have no evidence and no information that
this was the Russian government.”
Still, Kurm is pretty hopeless of ever gaining any further information that could be
legally useful for prosecuting anyone for cybercrimes against the Republic of Estonia.
In an interview in July 2007, he admitted to me: “The status is that we haven't got any
information from Russia and I'm quite sure that we will not get any information.”[16]
On January 25, 2008, Dmitri Galushkevich pled guilty to attacking Estonian

websites. He had to pay a fine of 17,500 Estonian kroons, or around $1,700 and
received only probation – no jail time. The case was closed, and no further legal action
was taken against anyone, largely because, in the words of Kurm, “Russia refused to
co-operate.”[17]
What this means, is that for the foreseeable future, cyberattacks will remain an
effective tactic countries between nations that are not exactly always friendly with one
another, as is the case with Russia and many of its former Soviet satellites.
7. Suggestions for Researchers and Policymakers to Improve Media Coverage
In summation, there are three main points that I would like researchers, policymakers
and journalists to come away with.
First, tone down the rhetoric, hyperbole, and watch your language. If you talk
about “cyberwar,” – the use of the word war has a very specific meaning and very
specific consequences. A war usually implies two, more-or-less equal sides, with a
clear objective. Cyberattacks generally are not always necessarily couched in the
applications of political conflict – in fact, many attacks have more to do with organized
crime or online mischief than they do actual warfare. As such, journalists should be
wary of sources that compare cyberattacks to nuclear warfare and make similarly
absurd comparisons. Further, researchers and policymakers need to be aware of the
words that they use themselves.
Second, researchers and policymakers need to be more open (as much as possible)
with the information that they do have. Journalists need to be able to verify data, and
understand the data that they're looking at. When everything is construed as a
“cyberwar,” it's tough to determine how various “cyberwars” compare to one another.
Was Estonia's attack the same as the one against Georgia? What about the 2009 attack
against the United States and South Korea?
Third, and most importantly, policymakers and researchers need to understand
how they can work together. Whether they like it or not, media can have a significant
influence on public policy. It is the job of the media to inform the public and act as a
watchdog on government's activities. The more information that public officials,
corporations and researchers can provide to journalists, the better the journalists can do
in presenting the case. However, one of the problems is that there simply aren't very
many journalists that fully understand neither how cyberattacks work nor what they
are. It would be helpful for journalists to participate in a workshop on cyberwarfare
from their local governments, or perhaps from the CCDCOE to better understand how
these attacks work from a technical standpoint.
References
[1] Rhoads, Christopher, “Kyrgyzstan Knocked Offline,” The Wall Street Journal. , January 28 2009.
http://online.wsj.com/article/SB123310906904622741.html?pagewanted=print
[2] Mackey, Robert, “Are 'Cyber-Militas Attacking Kyrgyzstan?', The New York Times
[3] Led Blog, February 5 2009. http://thelede.blogs.nytimes.com/2009/02/05/are-cyber-militias-
attacking-kyrgyzstan/?pagewanted=print
[4] “Severed cable disrupts net access,” BBC News, December 19, 2008
http://news.bbc.co.uk/2/hi/technology/7792688.stm
[5] “Georgia States Computers Hit By Cyberattack,” The Wall Street Journal, August 12, 2008.
http://online.wsj.com/article/SB121850756472932159.html
[6] Morozov, Evgeny, “An Army of Ones and Zeroes,” Slate, August 14, 2008.
http://www.slate.com/id/2197514/pagenum/all/#p2
[7] Landler, Mark, and Markoff, John. “Digital Fears Emerge After Data Siege in Estonia,” The New
York Times, May 29, 2007.
http://www.nytimes.com/2007/05/29/technology/29estonia.html?pagewanted=print
[8] “Estonia hit by 'Moscow cyber war,' BBC News, May 17, 2007.
http://news.bbc.co.uk/2/hi/europe/6665145.stm
[9] Traynor, Ian, “Russia accused of unleashing cyberwar to disable Estonia,” The Guardian, May 17,
2007. http://www.guardian.co.uk/world/2007/may/17/topstories3.russia
[10] Davis, Joshua, “Hackers Take Down the Most Wired Country in Europe,” Wired, August 21,
2007. http://www.wired.com/politics/security/magazine/15-09/ff_estonia?currentPage=all
[11] Farivar, Cyrus, “Cyberwar I,” Slate, May 22, 2007. http://www.slate.com/id/2166749/fr/flyout
[12] Davis, Joshua, “Hackers Take Down the Most Wired Country in Europe,” Wired, August 21,
2007. http://www.wired.com/politics/security/magazine/15-09/ff_estonia?currentPage=all
[13] Finn, Peter, “Cyber Assaults on Estonia Typify a New Battle Tactic,” The Washington Post, May
19, 2007.
http://www.washingtonpost.com/wp-dyn/content/article/2007/05/18/AR2007051802122.html
[14] “Remarks by the President in Photo Opportunity with Leaders of High-Tech Industry and Experts
on Computer Security,” The White House Office of the Press Secretary, February 15, 2000.
http://www.fas.org/irp/news/2000/02/000215-secure-wh1.htm
[15] “Cyber-attacks batter Web heavyweights,” CNN, February 9, 2000.
http://archives.cnn.com/2000/TECH/computing/02/09/cyber.attacks.01/index.html
[16] “Tulemused – Teksid,”
http://www.legaltext.ee/et/andmebaas/tekst.asp?loc=text&dok=X30068K7&k
eel=en&pg=1&ptyyp=RT&tyyp=X&query=karistusseadustik
[17] Margus Kurm, in discussion with the author, July 19 2007. Email from
Margus Kurm to the author, January 29 2008.
Part II
Technical Challenges and Solutions
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-191
Behavioral Analysis of Zombie Armies

Olivier THONNARD a,1 , Wim MEES a and Marc DACIER b
a
Royal Military Academy, Polytechnic Faculty, Brussels
b
Symantec Research Labs, Sophia Antipolis, France
Abstract. Zombie armies - or botnets, i.e., large groups of compromised machines

controlled remotely by a same entity - pose today a significant threat to national
security. Recent cyber-conficts have indeed demonstrated that botnets can be easily
turned into digital weapons, which can be used by cybercriminals to attack the net-
work resources of a country by performing simple Distributed Denial-of Service
(DDoS) attacks against critical web services. A deep understanding of the long-
term behavior of botnet armies, and their strategic evolution, is thus a vital require-
ment to combat effectively those latent threats. In this paper, we show how to en-
able such a long-term, strategic analysis, and how to study the dynamic behaviors
and the global characteristics of these complex, large-scale phenomena by apply-
ing different techniques from the area of knowledge discovery on attack traces col-
lected on the Internet. We illustrate our method with some experimental results ob-
tained from a set of worldwide distributed server honeypots, which have monitored
attack activity in 18 different IP subnets for more than 640 days. Our preliminary
results highlight several interesting findings, such as i) the strong resilience of zom-
bie armies on the Internet, with survival times going up to several months; ii) the
high degree of coordination among zombies; iii) the highly uneven spatial distribu-
tion of bots in a limited number of “unclean networks”, and iv) the large propor-
tion of home users’ machines with high-speed Internet connexions among the bot
population.
Keywords. Intelligence monitoring, Threat analysis, Zombie armies.
Introduction
In the recent years, many security experts have drawn attention to the increasingly im-
portant security problem related to zombie armies - also called botnets, which are groups
of malware-infected machines that are remotely controlled and coordinated by a same
entity. Still today, zombie armies and botnets constitute, admittedly, one of the main
threats on the Internet, as they are used for different kinds of illegal activities (e.g., bulk
spam sending, online fraud, denial of service attack, etc) [2,19]. More importantly, the
analysis of recent “cyber conflicts”, such as the presumed cases related to Estonia and
Georgia [17,6,7], have lead experts to the conclusion that botnets can be easily turned
into digital weapons, which can be used by cybercriminals (or dissidents) to attack the
network resources of a country by performing very simple Distributed Denial-of Service
(DDoS) attacks against critical web services (e.g., DNS servers, network routers, gov-
1 Corresponding Author: Olivier Thonnard, Royal Military Academy, Avenue de la Renaissance 30, 1000
Brussels, Belgium; E-mail: olivier.thonnard@rma.ac.be.

192 O. Thonnard et al. / Behavioral Analysis of Zombie Armies
ernment or financial websites, etc), which can lead to substantial economical or financial
loss. Although no clear evidence of the implication of any governmental organization in
those attacks could be underlined, one important lesson learned from these events is that
botnets are primarily used by dissidents or activists to perform this type of attacks in pe-
riods of political disturbances. A deep understanding of the long-term behavior of botnet
armies, and their evolution, is thus a vital requirement to be able to combat effectively
those latent threats.
While most previous studies related to botnets have focused on understanding their
inner working [24,5,1], or on techniques for detecting individual bots at the network-
level [8,9], in this work we are more interested in studying the global behaviors of those
armies from a strategic viewpoint. That is, we are not interested in studying a particular
botnet from the inside, or in the analysis of the various protocols used by bots to com-
municate with their C&C server. But instead, we want to perform a long-term, strategic
analysis of those armies from a behavioral point of view, i.e.: how long do they stay
alive on the Internet, what is their average size and their spatial distribution, and more
importantly, how do they evolve over time with respect to different criteria such as their
origins, or the type of activities (or scanning) they perform.
The first contribution of this paper consists in introducing a systematic method that
enables us to perform such a strategic analysis of zombie armies, based on the botnet
scanning traffic observed in a global honeynet. Our approach is based on an appropri-
ate combination of different knowledge discovery and data mining techniques, which
consists of the following components:
1. detection and characterization of coordinated attack events;
2. unsupervised clique-based clustering, so as to discover correlations among attack
events;
3. dimensionality reduction techniques, which allow us to visualize and to assess
the cliques correlations;
4. a fuzzy, multi-criteria decision-making process that leverages the results obtained
in the previous steps, in order to identify sequences of attack events that are very
likely attributed to the same zombie army.
As second contribution, we present some preliminary results obtained from a proof-
of-concept framework in which we implemented the techniques mentioned here above.
The experiments have been performed on attack traces collected with a worldwide dis-
tributed honeynet, which has observed global attack activity in over 18 different IP sub-
nets from Sep 2006 until July 2008 (i.e., about 640 days). Our experimental results high-
light several interesting facets of the botnet phenomenon:
• with a mean lifetime of about 98 days, zombie armies seem to be quite resilient.
In some extreme cases, we observed certain armies surviving for more than 18
months, which indicates that taking down botnets still constitutes a real challenge.
On average, zombie armies had at least 8,500 distinct, observable sources during
their lifetime.
• regarding the origins, malicious sources involved in zombie armies seem to be
highly unevenly distributed in the IPv4 address space; they clearly form a rel-
atively small number of tight clusters within a number of “unclean networks”,
which are thus responsible for a large deal of malicious activities related to server-
side attacks (e.g., network scanning, bot propagation).
O. Thonnard et al. / Behavioral Analysis of Zombie Armies 193
• over all zombie armies observed so far, at least 43% of the botnet population is
made of home users’ machines with high-speed Internet connexions (cable, DSL).
Windows 2000 and WinXP Pro were the primarily operating systems among zom-
bie machines (i.e., more than 90% of the bots).
• similarly to real-world armies, certain groups of zombie machines seem to be able
to coordinate their efforts, e.g., by coordinating different tasks such as network
reconnaissance and subsequent targeted attacks.
• finally, most of the identified zombie armies had a significant attack capability,
not only in terms of the available bandwidth that can possibly be offered by all
zombies together, but also the number of ports they are able to probe or to exploit.
The rest of the paper is structured as follows: in Section 1, we give a brief overview
of the honeynet used in our experiments, and we define the notion of coordinated at-
tack events as observed by the honeypots. In Section 2, we describe the components of
our knowledge discovery framework that we use to identify global attack phenomena,
whereof most are related to some activities of zombie armies. In Section 3, we present
our experimental results and the kind of findings we can obtain by applying this method
to a set of attack events collected on the Internet. Finally, we conclude in Section 4.
Note that this research builds on prior work in malicious traffic analysis. More par-
ticularly, we have presented in [28] a more formal and complete discussion of our frame-
work, especially regarding the aspect fuzzy, multi-criteria decision-making. To make this
paper as self-contained as possible, we have summarized as much as possible our pre-
vious contributions in Section 2. This paper will mostly focus on the practical results
obtained in each step of our analysis framework, rather than the formal aspects of the
different techniques.
1. Collecting Attack Traces with a Global Honeynet
1.1. Leurre.com Honeynet - Dataset Overview
Our data set is made of network attack traces collected with a distributed set of sen-
sors (called server honeypots), which are deployed in the context of the Leurre.com
Project [14,22]. Because honeypots are systems deployed for the sole purpose of being
probed or compromised, any network connection that they establish with a remote IP can
be considered as malicious, or at least suspicious.
Launched in 2003 by Eurecom, a research Institute based in Sophia Antipolis (France),
this project maintains a worldwide distributed system of honeypots running in more than
30 different countries covering the five continents. The main objective of the project is
to get a realistic picture of certain classes of global attack phenomena happening on the
Internet, by collecting unbiased quantitative data in a long-term perspective. In the first
phase of the project, the data collection infrastructure relied solely on low-interaction
sensors based on Honeyd [23] to collect unsolicited traffic (also sometimes termed “In-
ternet background radiation” [18]). In early 2008, a second phase of the project was
started with the deployment of medium-interaction honeypots based on the ScriptGen
[15] technology, in order to enrich the network conversations with the attackers. Script-
gen sensors are able to automatically learn about new protocol interactions, such that
they can handle 0-day exploits, and eventually capture shellcode samples and malware
Table 1. Overview of some prevalent types of activities observed in the honeynet, grouped by port sequence.
The network traffic has been collected from Sep’06 until June’08.
Observed Port Sequence Targeted Service Volume of Sources (%) Main Origins (countries)
|I ICMP (Echo request/reply) 755,227 (28%) US(20%),KR(11%),CN(10%),BR(6%),
others(53%)
|1026U|1027U|1028U Windows Messenger 373,361 (14%) CA(100%)
|1026U Windows Messenger 216,040 (8%) US(50%),null(17%),CA(6%), others(27%)
|445T Microsoft-DS 208,060 (8%) CS(32%),RS(19%),US(6%), others(43%)
|I|139T, |I|139T|445T ICMP (Allaple), MS-Netbios-ssn, Microsoft-DS 130,392 (5%) KR(20%), others(80%)
|135T Microsoft DCE/RPC 112,764 (4%) JP(16%),US(13%),CS(7%),RS(7%),
PL(6%),DE(6%), others(45%)
|5900T VNC 104,238 (4%) US(17%),CN(6%),FR(6%),KR(6%),
others(51%)
|2967T Symantec AntiVirus (ssc-agent) 101,062 (4%) US(23%),CN(8%),JP(6%),
DE(5%),PK(5%), others(53%)
|1433T MS-SQL 87,332 (3%) CN(32%),US(15%),others(53%)
|139T MS-Netbios-ssn 50,781 (2%) US(17%),CA(8%),TW(5%),FR(5%),
others(65%)
|I|80T ICMP, Web 48,649 (2%) US(54%),KR(11%),CN(8%),
CA(7%), others(20%)
|1434U MS-SQL-Monitor (Slammer) 36,627 (1%) CN(44%),US(14%),JP(6%), others(36%)
|22T SSH 36,094 (1%) CN(24%),US(13%),KR(8%),
TW(5%), others(50%)
|80T Web 28,005 (1%) US(27%),CN(7%),FR(7%),
DE(7%),null(5%), others(47%)
|137U MS-Netbios-ns 25,630 (<1%) US(16%),BR(9%),AR(6%),
FR(5%),ES(5%), others(59%)
|I|445T ICMP, Microsoft-DS 18,273 (<1%) US(14%),CN(13%),TW(8%),FR(7%),
JP(7%),null(6%),DE(5%), others(41%)
|4899T Remote Admin 15,935 (<1%) CN(15%),US(15%),KR(10%),
RU(5%), others(54%)
binaries when they are targeted by code injection attacks. All network traces captured on
the platforms are automatically uploaded into a centralized database. The collected traf-
fic is also enriched with a diverse set of contextual information, such as: the geograph-
ical location and the ISP’s of malicious sources (via Maxmind), reverse DNS lookups,
VirusTotal2 and Anubis3 reports for each sample of downloaded malware, passive OS
fingerprinting (with P0f), Snort IDS alerts, and more recently, we also added the correla-
tion of the observed IP sources with different IP blacklisting services (e.g., Spamhaus4 ,
Emergingthreats5 blocking lists, and a fast-flux bot tracker6 ).
For the purpose of this study, we have used a 640-day attack trace collected by 36
platforms located in 20 different countries and belonging to 18 different class A-subnets.
Note that, in the scope of this paper, we only considered the traffic collected by low-
interaction sensors; but we are actively looking into extending our analysis techniques
to integrate the attack traffic gathered by the medium-interaction (ScriptGen) platforms.
Table 1 gives an overview of the most prevalent types of activities grouped by targeted
port sequences, and their origins, as observed in the honeynet.
From this traffic, we have then selected only the most prevalent types of activities
observed on the sensors, i.e., about 130 distinct attack profiles for which an activity
2 http://www.virustotal.com
3 http://anubis.iseclab.org
4 http://www.spamhaus.org
5 http://www.emergingthreats.net
6 http://dnsbl.abuse.ch
Figure 1. Distribution of malicious sources in the IPv4 address space using a fractal mapping (Hilbert curve).
involving a sufficient number of IP sources had been observed at least once on a given
day. This data set comprises totally 1,195,254 distinct sources, which have sent about
3,423,577 packets to the sensors. Fig. 1 illustrates the distribution of malicious sources
for these activities using a fractal mapping (e.g., a Hilbert curve). Note that spoofed
IP addresses have already been filtered from this data set. As such, Fig. 1 and Table 1
give already some interesting viewpoints, as it clearly shows that most malicious sources
seem to be clustered in a limited number of IP blocks (or AS’es). Nevertheless, this type
of global analysis does not help us to get insights into the individual attack phenomena
that occurred at a large scale (such as zombie armies). Moreover, such global trends do
not allow us to learn about the modus operandi of the attackers, which is why we need
to develop a more detailed analysis.
1.2. Coordinated Attack Events
We use a classical clustering algorithm to perform a first low-level classification of the

raw network traffic. Hence, each IP source observed on a honeypot sensor is attributed
to a so-called attack cluster [21] according to its network characteristics, such as the
number of IP addresses targeted on the sensor, the number of packets and bytes sent to
each IP, the attack duration, the average inter-arrival time between packets, the associated
port sequence being probed (e.g., if a source sends first some ICMP packets followed
by an exploit on port 445/TCP, then it is associated to the port sequence I-445T), and
the packet payload. Therefore, all IP sources belonging to a given attack cluster have left
450
AE103 (139T) on sensor 45
400 AE171 (1433T) on sensor 45
AE173 (5900T) on sensor 9
350
300
Nr of sources
250
200
150
100
50
0
82 84 86 88 90 92 94 96 98 100 102
Time (by day)
Figure 2. Illustration of 3 attack events observed on 2 different sensors, and targeting 3 different ports.
very similar network traces on a given sensor and consequently, they can be considered
as having the same attack profile. This leads us then to the concept of attack event, which
is defined as follows:
An attack event refers to a subset of IP sources having the same attack profile on a given
sensor, and whose coordinated activity has been observed within a specific time window.
Fig. 2 illustrates this notion by representing the time series (i.e., the number of
sources per day) of three coordinated attack events observed on two different sensors
in the same time interval, and targeting three different ports. The identification of those
events can be easily automated by using the method presented in [20]. By doing so, we
are able to extract interesting events from the spurious, nonproductive traffic collected
by our sensors, and we can focus on the most important events that might originate
from coordinated phenomena, such as attack activities resulting from botnet reconnais-
sance scans, and bot propagation. As previous botnet studies have already showed [13],
it seems that the botnet scanning behavior is ingrained to the botnets because this is an
effective (and low-cost) way for them to recruit new bots. Therefore, botmasters will
probably not give up scanning in the near future.
By using the technique described in [20], we have extracted from the whole data
set about 351 attack events that were coordinated on at least two different sensors. In
the rest of this paper, we will focus on the analysis of this set of attack events, which
still accounts for 282,363 unique sources (23.6 % of the original data set), or 741,349
packets (21.5%), and we will show how to take advantage of different external attack
characteristics to discover knowledge, and to identify individual phenomena related to
zombie armies.
2. A Framework to Identify Global Attack Phenomena
2.1. Overview
Once we have identified a set of attack events occurring at different moments, how could
we know in a reliable way which events can be attributed to the same root phenomenon?
That is, how can we identify which sequences of attack events are very likely the conse-
quence of the same zombie army scanning or probing one or several subnets, eventually
during non-contiguous intervals of time?
In the realm of threat monitoring, this problem is sometimes referred to as “attack

attribution”, which is the process of effectively attributing new attack events to (un)-
known phenomena, based on some evidence or traces left on one or several monitoring
platforms. To address this problem in a systematic way, we have developed a framework
that analyzes attack events with appropriate knowledge discovery (KDD) techniques.
The main components of this framework are sketched in Fig. 3. Based on a set of attack
events (as defined here above), the first KDD component extracts cliques of attackers
in an unsupervised way, so as to identify meaningful correlations between events. That
is, we want to know whether some groups of events are strongly correlated with respect
to some given characteristics. For example, we could discover which groups of attack
events share the very same spatial distributions (in terms of geographical or IP subnet
distributions), or which other groups of attack events are targeting the same set of sensors
in the same time interval, or which groups of attacks are similar in terms of activities
(e.g., the port sequences targeted by malicious sources), and so on. We motivate our
choice of attack characteristics used to discover knowledge in the next subsection. Then,
we evaluate the consistency of the extracted cliques (or clusters) by using dimensional-
ity reduction techniques, which enable us to visualize on a map the cliques results for
each attack dimension. We refer to this step as “semantic mapping”, since the distance
between each pair of events on a given mapping has a certain meaning. Indeed, the dis-
tances are related to the degree of similarity between the underlying feature vectors of
the attack events (i.e., the distributions of countries, subnets, etc).
In the next component of the framework, we have implemented a multi-criteria

decision-making algorithm that is based on fuzzy inference systems (FIS). The objective
consists in combining intelligently the previously extracted knowledge (i.e., the cliques
and the semantic mappings), so as to build sequences of attack events that can be at-
tributed to the same global phenomena with a high degree of confidence, thanks to the
combination of different statistical measurements. Interestingly, a FIS does not need any
training prior making inferences. Instead, it takes only advantage of the previously ex-
tracted knowledge to make sound inferences, so as to attribute incoming attack events to
a given phenomenon. Each identified attack phenomenon is then modeled with a fuzzy
inference system.
2.2. Defining Attack Characteristics
In most knowledge discovery applications, we must first define salient features that may
provide some meaningful patterns [11]. So, we start by defining different attack char-
Figure 3. Components of a Knowledge Discovery Framework for Identifying Global Phenomena.
acteristics that we have used to extract knowledge from our set of attack events. In this
specific case, we consider them as useful to analyze the root causes of global phenomena
observed on our sensors, and as a result, to identify different zombie armies. However,
we do not pretend they are the only ones that could be used in threat monitoring. Since
other characteristics might prove relevant in the future, our framework is built such that
additional features could be easily included when necessary (e.g., to include characteris-
tics related to code injection attacks, shellcodes, or malware samples).
The two first characteristics retained are related to the origins of the attackers, i.e.
their spatial distributions. First, the geographical distribution of malicious sources can
be used to identify botnets that are located in a limited number of countries. Similarly,
the IP network blocks provide also an interesting viewpoint on the attack phenomena,
since it gives a good indication of the spatial “uncleanliness” of certain networks, i.e., the
tendency for compromised hosts (e.g., zombie machines) to stay clustered within unclean
networks [4]. So, for each attack event, we can create a feature vector representing either
the distribution of originating countries, or of IP addresses grouped by Class A-subnet
(i.e., by /8 prefix).
The next characteristic deals with the targets of the attackers, namely the distribu-
tion of sensors that have been targeted by the sources. Botmasters may indeed send com-
mands at a given time to all zombies to instruct them to start scanning (or attacking) one
or several IP subnets, which of course will create coordinated attack events on specific
sensors. Therefore, it seems important to look at relationships that may exist between
attack events and the sensors they have been observed on.
Besides the origins and the targets, the type of activity performed by the attackers
seems also relevant to us. In fact, bot software is often crafted with a certain number of
available exploits targeting a reduced set of TCP or UDP ports. In other words, we might
think of each botnet having its own attack capability, which means that a botmaster will
normally issue scan or attack commands only for vulnerabilities that he might exploit to
expand his botnet. So, it seems to make sense to take advantage of this feature to look
for similarities between the sequences of ports that have been targeted by the sources of
the attack events.
Table 2. Some experimental clique results obtained from a honeynet dataset collected from Sep 06 until June
08. (1) the given patterns represent the average distributions for the most prevalent cliques, i.e. the ones lying
in the upper quartile in terms of number of sources. For the IP subnets (resp. targeted platforms), the numbers
refer to the distributions of originating (resp. targeted) class A-subnets.
Attack Dimension Nr of Max.size Min.size Volume of Most prevalent patterns found in the cliques(1)
Cliques (nr events) (nr events) sources (%)
Geolocation 31 40 3 84.4 CN,CA,US,FR,TW, IT,ES,FR,SE,DE,IL, KR,US,BR,PL,CN,CA
US,JP,GB,DE,CA,FR,CN,KR, US,FR,JP,CN,DE,ES,TW, CA,CN
PL,DE,ES,HU,FR
IP Subnets (Class A) 25 51 3 91.2 87,82,151,83,84,81,85,213, 222,221,60,218,58,24,124,121,219,82,220
201,83,200,24,211,218,89,124,61,82,84, 24,60
83,84,85,80,88, 193,195,201,202,203,216,200,61,24,84,59
Targeted platforms 17 86 2 70.1 202, 88, 192, 195, 193, 194
129, 134, 139, 150, 24, 213
Port sequences 22 66 4 93.2 I, 1433T, I-445T, 5900T, 1026U, 135T, 50286T
I-445T-139T-445T-139T-445T, 6769T, 1028U-1027U-1026U
Finally, we have also decided to compute, for each pair of events, the ratio of com-
mon IP addresses. We are aware of the fact that, as time passes, some zombie machines
of a given botnet might be cured while others may get infected and join the botnet. Addi-
tionally, certain ISPs apply a quite dynamic policy of IP address allocation to residential
users, which means that bot-infected machines can have different IP addresses when we
observe them at different moments (i.e., DHCP churn effect). Nevertheless, and accord-
ing to our domain experience, it is reasonable to expect that if two distinct attack events
have a high percentage of IP addresses in common, then the probability that those two
events are somehow related to the same global phenomenon is increased (assuming that
the time difference between the two events is not too large).
2.3. Clique-based Knowledge Discovery
For each attack characteristic considered here above, we have applied a clique-based
clustering on our set of attack events. That is, we use a graph-based approach to formu-
late the problem: the vertices of the graph represent the feature vectors of each attack
event (e.g., the distribution of countries, subnets, targeted sensors, etc), and the edges
express the similarity relationships between those vertices. Clearly, the choice of a simi-
larity metric is very important, as it has an impact on the properties of the final clusters,
such as their size, quality, and consistency. To reliably compare the kind of empirical dis-
tributions mentioned here above, we have chosen to rely on strong statistical distances,
such as Pearson’s χ2 , or the Jensen-Shannon divergence (JSD) [16], which derives itself
from the Kullback-Leibler divergence [12]. Finally, the clustering is performed by ex-
tracting so-called maximal weighted cliques (MWC) from the graph, where a maximal
clique is defined as an induced sub-graph in which the vertices are fully connected and
it is not contained within any other clique. We refer the interested reader to [27,26] for a
more detailed description of this clique-based clustering technique applied to honeynet
traces.
Table 2 presents a high-level overview of the cliques obtained for each attack di-
mension separately. As we can see, a relatively high volume of sources could be clas-
sified into cliques for each dimension. The last colon with the most prevalent patterns
gives an indication of which countries or class A-subnets (e.g., originating or targeted IP
subnets) are most commonly observed in the cliques that lie in the upper quartile with
respect to the number of sources. Interestingly, it seems that many coordinated attack
events are coming from a given IP sub-space. Regarding the targeted platforms, several
cliques involve a single class A-subnet. About the type of activities, we can observe some
commonly targeted ports (e.g., Windows ports used for SMB or RPC, or SQL and VNC
ports), but also a large number of uncommon high TCP ports that are normally unused on
standard (and clean) machines (such as 6769T, 50286T, 9661T, . . . ). A non-negligeable
volume of sources is also due to UDP spammers targeting Windows Messenger popup
service (ports 1026 to 1028/UDP).
2.4. Visualizing Cliques - Knowledge Consolidation
In order to assess the consistency of the resulting cliques of attack events, it can be useful
to see them charted on a two-dimensional map so as to i) verify the proximities among
clique members (intra-clique consistency), and ii) understand potential relationships be-
tween different cliques that are somehow related (i.e. inter-clique relationships). More-
over, the statistical distances used to compute those cliques make them intrinsically co-
herent, which means also that certain cliques of events may be somehow related to each
other, although they were separated by the clique algorithm.
Since most of the feature vectors we are dealing with have a high number of variables
(e.g., a geographical vector has more than 200 country variables), the structure of such
high-dimensional data set cannot be displayed directly on a 2D map. Multidimensional
scaling (MDS) is a set of methods that can help to address this problem. MDS is based on
dimensionality reduction techniques, which aim at converting a high-dimensional dataset
into a two or three-dimensional representation that can be displayed, for example, in a
scatter plot. The aim of dimensionality reduction is to preserve as much of the significant
structure of the high-dimensional data as possible in the low-dimensional map. As a
consequence, MDS allows an analyst to visualize how far observations are from each
other for different kinds of similarity measures, which in turn can deliver insights into
the underlying structure of the high-dimensional dataset.
Because of the intrinsic non-linearity of real-world data sets, we have applied a
recent MDS technique called t-SNE to visualize each dimension of the data set, and
to assess the consistency of the cliques results. t-SNE [29] is a variation of Stochastic
Neighbour Embedding; it produces significantly better visualizations than other MDS
techniques by reducing the tendency to crowd points together in the centre of the map.
Moreover, this technique has proven to perform better in retaining both the local and
global structure of real, high-dimensional datasets in a single map, in comparison to other
non-linear dimensionality reduction techniques such as Sammon mapping, Isomaps or
Laplacian Eigenmaps [10].
Figure 4 shows the resulting two-dimensional plot obtained by mapping the geo-
graphical vectors on a 2D map using t-SNE. Each datapoint on this map represents the
geographical distribution of a given attack event. The coloring refers to the clique mem-
bership of each event, and the dotted circles indicate the clique sizes. We could easily
verify that two adjacent events on the map have highly similar geographical distributions
(even from a statistical viewpoint), while two distant events have clearly nothing in com-
mon in terms of originating countries. Quite surprisingly, the resulting mapping is far
from being “chaotic”; it presents a relatively sparse structure with clear datapoint group-
Figure 4. Visualization of geographical cliques of attackers. The coloring refers to the different cliques and the
dotted circles indicate their sizes on the low-D map. The superposed text labels indicate the two first attacking
countries of the distribution of certain attack events, as well as some of the targeted port sequences (in red).
ings, which means also that most of those attack events present very tight relationships
regarding their origins. Due to the strict statistical distances used to calculate cliques,
this kind of correlation can hardly be obtained by chance only.
Similar “semantic mapping” can naturally be obtained for the other dimensions (e.g.,
subnets, platforms, etc), so as to help assessing the quality of other cliques of attackers.
As described in the next Section, those different mappings will be used by the multi-
criteria decision-making component of our framework to identify global phenomena, i.e.
by combining efficiently different sets of cliques.
2.5. Identification of Zombie Armies using Fuzzy Inferences
The final objective consists in re-constructing sequences of attack events that can be at-
tributed with a high confidence to the same root phenomenon in function of multiple
criteria. In other words, we want to build an inference engine that takes as input the ex-
tracted knowledge (cliques and mappings) to classify incoming attack events into either
“known phenomena”, or otherwise to identify a new phenomenon when needed (e.g.,
when we observe a new zombie army). To do this, we have implemented a multi-criteria
decision-making algorithm that relies on fuzzy inferences. Our motivation is that: i) we
have a priori zero-knowledge of the expected output, which means that we can not pro-
vide training samples showing the characteristics of the output we are looking for; and ii)
we want to include some domain knowledge to specify which type of combinations we
expect to be promising in the root cause identification. Also, we favor the “white-box”
approach (or a transparent reasoning process), which allows an expert to understand why
the system has grouped a given set of events into the same root phenomenon.
Figure 5. Main components of a Fuzzy System.
Although large-scale phenomena on the Internet are complex and dynamic, our in-
tuition is that two consecutive attack events should be linked to the same root phe-
nomenon if and only if they share at least two different attack characteristics. That is,
our decision-making process will attribute two attack events to the same phenomenon
when the events characteristics are “close enough” (from a statistical viewpoint) for
any combination of at least two attack dimensions out of the complete set of crite-
ria: {origins, targets, activity, commonIP }. In other words, we hypothesize that real-
world phenomena may perfectly evolve over time, which means that two consecutive
attack events of the same zombie army must not necessarily have all their attributes in
common. For example, the bots’ composition of a zombie army may evolve over time
because of the cleaning of infected machines and the recruitment of new bots. From our
observation viewpoint, this will translate into a certain shift in the IP subnet distribution
of the zombie machines for subsequent attack events of this army (and thus, most proba-
bly different cliques w.r.t. the origins). Or, a zombie army may be instructed to scan sev-
eral consecutive IP subnets in a rather short interval of time, which will lead to the obser-
vation of different events having highly similar distributions of originating countries and
subnets, but those events will target completely different sensors, and may eventually use
different exploits (hence, targeting different port sequences).
On the other hand, we consider that only one correlated attack dimension is not suf-
ficient to link two attack events to the same root cause, since the result might then be due
to chance only (e.g., a large proportion of attacks originate from some large or popular
countries, certain Windows ports are commonly targeted, etc). However, by combining
intelligently several attack viewpoints, we can reduce considerably the probability that
two attack events would be attributed to the same root cause whereas they are in fact
unrelated.
We still need to formally define what is the “relatedness degree” between two attack
events, certainly when they do not belong to a same clique but are somehow “close” to
each other. Intuitively, attack events characteristics in the real world have unsharp bound-
aries, and the membership to a given phenomenon can be a matter of degree. For this
reason, we have developed a decision-making process that is based on a fuzzy inference
system (FIS). Fuzzy Inference is a convenient way to map an input space to an output
space with a flexible and extensible system, and using the codification of common sense
and expert knowledge. The mapping then provides a basis from which decisions can be
made. The main components of an inference system are sketched in Fig. 5. To map the
input space to the output space, the primary mechanism is a list of if-then statements
called rules, which are evaluated in parallel, so the order of the rules is unimportant.
Instead of using crisp variables, all inputs are fuzzified using membership functions in
order to determine the degree to which the input variables belong to each of the appro-
priate fuzzy sets. If the antecedent of a given rule has more than one part (i.e., multiple
’if’ statements), a fuzzy logical operator is applied to obtain one number that represents
the result of the antecedent for that rule.
Concretely, we use the knowledge obtained from the extraction of cliques to build
the fuzzy rules that describe the behavior of a given phenomenon. The characteristics
of new incoming attack events are then used as input to the fuzzy systems that model
the phenomena identified so far. In each of those fuzzy systems, the features of the most
recent attack event shall define the current parameters of the membership function used
to evaluate the following simple rules: if xi is close AND if yi is close then zi is related,
∀i ∈ {geo, subnets, targets, portsequence}. The membership functions referred to as
“is close” in the fuzzy rules are thus defined by the characteristics of the cliques to
which the attack events belong. The calculation of the rule output zi ∈ [0, 1] is just
the intersection between two curves, which quantifies the inter-relationship between the
cliques (and hence, between the attack events).
The results of all rules are then combined and distilled into a single, crisp value
using an appropriate multi-criteria aggregation function. In this case, we use an Ordered
Weighted Average (OWA) operator, which allows to model more complex requirements
such as “most of”, or “at least two” criteria to be satisfied in the overall decision function
[30]. We refer the interested reader to [28] for a more detailed discussion of our multi-
criteria decision-making algorithm.
3. Behavioral Analysis of Zombie Armies
3.1. Global Characteristics
In this Section, we provide some experimental results obtained by applying our multi-
criteria inference method to our set of attack events introduced in Section 2 (clique anal-
ysis). Over the whole collection period (640 days), we found only 32 global phenomena.
In total, 348 attack events (99%) could be attributed to a large-scale phenomenon. An
in-depth analysis has revealed that most of those phenomena (apart from the noisy net-
work worm W32.Rahack.H [25], also known as W32/Allaple) are quite likely related to
zombie armies, i.e. groups of compromised machines belonging to the same botnet(s).
We conjecture this for the following main reasons: i) the apparent coordination of the
sources, both in time (i.e., coordinated events on several sensors) and in the distribution
of tasks (e.g., scanners versus attackers); ii) the short durations of the attack events, typ-
ically a few days only, whereas “classical” worms tend to spread over longer, contin-
uous periods of time; iii) the absence of known classical network worm spreading on
many of the observed port sequences; and iv) the source growing rate, which has a sort
of exponential shape for worms and is somehow different for botnets [13].
To illustrate the results, Table 3 presents an overview of some global phenomena
found in our dataset. Thanks to our method, we are able to characterize precisely the
behaviors of the identified phenomena or zombie armies. Hence, we found that the largest
army had in total 57 attack events comprising 69,884 sources, and could survive for
about 112 days. The longest lifetime of a zombie army observed so far was still 586
Total size (nr of sources)

0 10k 20k 30k 40k 50k 60k 70k
1 1
0.9 0.9
CDF size
0.8 0.8
CDF lifetime
0.7 0.7
0.6 0.6
F(x)
F(x)
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 100 200 300 400 500 600
Lifetime (nr of days)
Figure 6. Empirical CDF of the size and lifetime of zombie armies.
days. Fig. 6 shows the cumulative distributions (CDF) of the lifetime and size of the
identified armies. Those figures reveal some interesting aspects of their global behaviors:
according to our observations, at least 20% of the zombie armies had in total more than
ten thousand observable7 sources during their lifetime, and the same proportion of armies
could survive on the Internet for at least 250 days. On average, zombie armies have a
total size of about 8,500 observed sources, a mean number of 658 sources per event, and
their mean survival time is 98 days.
Regarding the origins, we observe some very persistent groups of IP subnets and
countries of origin across many different armies. On Fig. 7, we can see the CDF of the
sources involved in the zombie armies of Table 3, where the x-axis represents the first
byte of the IPv4 address space. It appears clearly that malicious sources involved in those
phenomena are highly unevenly distributed and form a relatively small number of tight
clusters, which account for a significant number of sources and are thus responsible for
a large deal of the observed malicious activities. This is consistent with other prior work
on monitoring global malicious activities, in particular with previous studies related to
measurements of Internet background radiation [3,18,31]. However, we are now able to
show that there are still some notable differences in the spatial distributions of those
zombie armies with respect to the average distribution over all sources (represented with
the blue dashed line). In other words, certain armies of compromised machines can have
very different spatial distributions, even though there is a large overlap between “zombie-
friendly” IP subnets. Moreover, because of the dynamics of this kind of phenomena,
we can even observe very different spatial distributions within a same army at different
moments of its lifetime. This is a strong advantage of our analysis method that is more
precise and enables us to distinguish individual phenomena, instead of global trends, and
to follow their dynamic behavior over time.
7 It is important to note that the sizes of the zombie armies given here only reflect the number of sources we
could observe on our sensors; the actual sizes of those armies are most probably much larger.
Figure 7. Empirical CDF of sources in IPv4 address space for the 9 zombie armies illustrated in Table 3.
Another interesting observation on Fig. 7 is related to the subnet CDF of ZA1 (uni-
formly distributed in the IPv4 space, which means randomly chosen source addresses)
and ZA20 (a constant distribution coming exclusively from the subnet 24.0.0.0/8). A very
likely explanation is that those zombie armies have used spoofed addresses to send UDP
spam messages to the Windows Messenger service. So, this indicates that IP spoofing is
still possible under the current state of filtering policies implemented by certain ISP’s on
the Internet.
Then, in terms of attack capability, we observe that about 50% of the armies could
target at least two completely different ports (thus, probably two different exploits, at
least), and one army had even an attack capability greater than 10. Table 4 provides addi-
tional details on the characteristics of malicious sources involved in those zombie armies.
Regarding the operating systems (detected through passive OS fingerprinting with P0f),
we can see that a large majority of the sources are running either Windows 2000 SP or
Windows XP Pro. Finally, by analyzing the hostnames of the sources (obtained via re-
verse DNS lookups), we infer the ratio of home users’s machines by looking for typical
strings such as ’%DSL%’, ’%PPP%’, ’%CABLE%’. Over all zombie armies observed
so far, we found that at least 43% of the botnet population is made of residential users
with high-speed Internet connections. If we take 256kbps as a lower-bound estimate of
the average upstream bandwidth for this kind of connection, then we observe that most
of those zombie armies could have an aggregate network capacity of several gigabits per
seconds, which can easily be used to exhaust almost any type of network resources on
the Internet by launching Distributed Denial of Service attacks.
3.2. Some Detailed Examples
In this Section, we further detail two zombie armies to illustrate some typical behaviors
we could observe among the identified phenomena, e.g.:
i) a move (or drift) in the origins of certain armies (both geographical and IP blocks)
during their lifetime;
Table 3. Overview of some large-scale phenomena found in a honeynet dataset (Sep’06 until Jun’08.
Id Nr of Total size Lifetime Targeted sensors Attack capability Main origins

events (nr sources) (nr days) (Class A- subnets) (countries / subnets)
1 10 18,468 535 24.*,193.*,195.*,213.* 1026U US,JP,GB,DE,CA,FR,CN,KR,NL,IT
69,128,195,60,81,214,211,132,87,63
4 82 26,962 321 202.* 12293T,15264T,18462T,25083T,
25618T,28238T,29188T, IT,ES,DE,FR,IL,SE,PL
32878T,33018T,38009T,4152T, 87,82,83,84,151,85,81,88,80
46030T,4662T,50286T,. . .
5 13 9,644 131 195.* 135T,139T,1433T,2968T,5900T CN,US,PL,IN,KR,JP,FR,MX,CA
218,61,222,83,195,221,202,24,219
6 15 51,598 >1 year > 7 subnets ICMP (W32.Rahack.H / Allaple) KR,US,BR,PL,CN,CA,FR,MX,TW
201,83,200,24,211,218,89,124
9 23 11,198 218 192.*,193.*,194.* 2967T,2968T,5900T US,CN,TW,FR,DE,CA,BR,IT,RU
193,200,24,71,70,213,216,66
10 57 69,884 112 128.*,129.*,134.*,139.*,150.* I-I445T CN,CA,US,FR,TW,IT,JP,DE
222,221,60,218,58,24,70,124
11 14 2,636 110 129.*,134.*,139.*,150.* I-445T-139T-445T-139T-445T US,FR,CA,TW,IT
82,71,24,70,68,88,87
12 14 27,442 183 192.*,193.*,194.*,195.* 1025T,1433T,2967T US,JP,CN,FR,TR,DE,KR,GB
218,125,88,222,24,60,220,85,82
20 10 30,435 337 24.*, 129.*, 195.* 1026U,1026U1028U1027U,1027U CA,CN
24,60
Table 4. Some detailed characteristics related to the composition of different zombie armies.
Zombie Army Id Home Users (DSL, Cable, PPP) Operating Systems (P0f)
1 spoofed IP’s -
4 69% Windows 2000 SP (68%), Windows XP Pro (5%)
6 38% Windows 2000 SP (2%), unknown (98%)
20 spoofed IP’s -
ii) a large scan sweep by the same army targeting several consecutive class A-
subnets;
iii) within a same army, multiple changes in the port sequences (or exploits) used
by zombies to scan or to attack;
iv) a coordination between different armies.
Zombie army 12 (ZA12) is an interesting case in which we can observe the behaviors
ii) and iii). Fig. 8 represents the output of the fuzzy system modeling this phenomenon.
Each bar graph represents the fuzzy output zi for a given attack dimension, whereas the
last plot shows the final aggregated output from which the decision to group those events
together was made (i.e., F (zi )). We can clearly see that the targets and the activities of
this army have evolved between certain attack events (e.g., when the value of zi is low).
That is, this army has been scanning (at least) four consecutive class A-subnets during
its lifetime (still 183 days), while probing at the same time three different ports on these
subnetworks.
Then, the largest zombie army observed by the sensors (ZA10) has showed the be-
haviors i) and iv). On Fig. 9, we can see that this army had four waves of activity during
which it was randomly scanning 5 different subnets (note the almost perfect coordination
among those attack events) on Windows ports (445T, 139T), preceded by ICMP. When
inspecting the subnet distributions of those different attack waves, we could clearly ob-
Geo
0.5
0
1 2 3 4 5 6 7 8 9 10 11 12 13
Port Seq Targets Subnets

1
0.5
0
1 2 3 4 5 6 7 8 9 10 11 12 13
1
0.5
0
1 2 3 4 5 6 7 8 9 10 11 12 13
1
0.5
0
1 2 3 4 5 6 7 8 9 10 11 12 13
Fuzzy Common
1
IPs
0.5
0
1 2 3 4 5 6 7 8 9 10 11 12 13
output
1
0.5
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13
Attack events (ordered in time)
Figure 8. Output of the fuzzy inference system (zi and F (zi )) modeling the zombie army nr 12.
1400
1200
1000
Nr of sources
800
600
400
200
0
100 120 140 160 180 200 220
Time (by day)
Figure 9. Time series of coordinated attack events for zombie army ZA10 (Nr of sources / day).
serve a drift in the origins of those sources, quite likely as certain machines were infected
by (resp. cleaned from) the bot software. Finally, we found another smaller army (ZA11)
that is clearly related to ZA10 (e.g., same temporal behavior, similar activity, same tar-
gets); but in this case, a different group of zombie machines, resulting in very different
subnet CDF’s on Fig. 7), was used to attack only specific IP addresses on our sensors,
probably by taking advantage of the results given by the army of scanners (ZA10). The
scanners were probably using some OS fingerprinting techniques to detect Windows op-
erating systems, since only those ones were targeted by the attackers on ports 445 and
Figure 10. Visualization of the distributions of subnets of origins for Zombie armies 10 and 11, which involve
two distinct communities of machines (scanners and attackers). The labels indicate the cliques’ memberships
of the attack events represented by the data points.
139 (and not the Linux honeypots). The distinction between scanners and attackers is
even more visible on the 2D mapping (illustrated on Fig 10) obtained from the subnets
distributions of these two zombie armies.
4. Conclusions
In this paper, we have introduced an analysis framework to identify, observe and char-
acterize zombie armies on the Internet, based on the attack traces they have left on dis-
tributed sensors. Recent cyber-conflicts have showed that zombie armies and botnets can
be easily turned into digital weapons and used to perform DDoS attacks against the net-
work infrastructure of a Nation. It is thus very important to understand the long-term be-
havior of botnet armies, and their strategic evolution, in order to deploy effective counter-
measures against those latent threats. Our analysis is based on the application of appro-
priate knowledge discovery techniques and a multi-criteria decision-making process. A
key aspect of the proposed method is the exploitation of external characteristics of mali-
cious sources, such as their spatial distributions in terms of countries and IP subnets. Our
experiments on a set of real-world attack traces have also highlighted some interesting
aspects of the global characteristics of such zombie armies, such as their high resilience
and the high attack capacity that zombie machines can potentially offer. As future work,
we envisage to extend our method to other data sets, such as high-interaction (client)
honeypot data, or malware data sets, and to include even more relevant attack features so
as to improve further the inference capabilities of the system, and thus also our insights
into malicious behaviors observed on the Internet.
Acknowledgements
This work has been partially supported by the European Commission through project
FP7-ICT-216026-WOMBAT funded by the 7th framework program. The opinions ex-
pressed in this paper are those of the authors and do not necessarily reflect the views of
the European Commission.
References
[1] Paul Barford and Vinod Yegneswaran. An Inside Look at Botnets. Advances in Information Security.
Springer, 2006.
[2] David Barroso. Botnets - the silent threat. In European Network and Information Security Agency
(ENISA), November 2007.
[3] Zesheng Chen, Chuanyi Ji, and Paul Barford. Spatial-temporal characteristics of internet malicious
sources. In Proceedings of INFOCOM, pages 2306–2314, 2008.
[4] M. P. Collins, T. J. Shimeall, S. Faber, J. Janies, R. Weaver, M. De Shon, and J. Kadane. Using un-
cleanliness to predict future botnet addresses. In IMC ’07: Proceedings of the 7th ACM SIGCOMM
conference on Internet measurement, pages 93–104, New York, NY, USA, 2007. ACM.
[5] Evan Cooke, Farnam Jahanian, and Danny McPherson. The Zombie roundup: Understanding, detecting,
and disrupting botnets. In Proceedings of the Steps to Reducing Unwanted Traffic on the Internet (SRUTI
2005 Workshop), Cambridge, MA, July 2005.
[6] Crime-Research. Cyberwar: Russia vs estonia, http://www.crime-research.org/articles/cyberwar-russia-
vs-estonia/, [may 09].
[7] Darkreading. Botnets behind georgian attacks offer clues, http://www.darkreading.com/security/app-
security/showarticle.jhtml?articleid=211201216, [may 09].
[8] G. Gu, R. Perdisci, J. Zhang, and W. Lee. BotMiner: Clustering analysis of network traffic for protocol-
and structure-independent botnet detection. In Proceedings of the 17th USENIX Security Symposium,
2008.
[9] Guofei Gu, Junjie Zhang, and Wenke Lee. BotSniffer: Detecting botnet command and control chan-
nels in network traffic. In Proceedings of the 15th Annual Network and Distributed System Security
Symposium (NDSS’08), February 2008.
[10] Geoffrey Hinton and Sam Roweis. Stochastic neighbor embedding. In Advances in Neural Information
Processing Systems 15, volume 15, pages 833–840, 2003.
[11] A.K. Jain and R.C. Dubes. Algorithms for Clustering Data. Prentice-Hall advanced reference series,
1988.
[12] S. Kullback and R. A. Leibler. On information and sufficiency. Annals of Mathematical Statistics 22:
79-86., 1951.
[13] Wenke Lee, Cliff Wang, and David Dagon, editors. Botnet Detection: Countering the Largest Security
Threat, volume 36 of Advances in Information Security. Springer, 2008.
[14] C. Leita, V.H. Pham, O. Thonnard, E. Ramirez-Silva, F. Pouget, E. Kirda, and Dacier M. The Leurre.com
Project: Collecting Internet Threats Information Using a Worldwide Distributed Honeynet. In Proceed-
ings of the WOMBAT Workshop on Information Security Threats Data Collection and Sharing, WIST-
DCS 2008. IEEE Computer Society press, April 2008.
[15] Corrado Leita, Ken Mermoud, and Marc Dacier. Scriptgen: an automated script generation tool for
honeyd. In Proceedings of the 21st Annual Computer Security Applications Conference, December
2005.
[16] J. Lin. Divergence measures based on the shannon entropy. Information Theory, IEEE Transactions on,
37(1):145–151, Jan 1991.
[17] Arbor Networks. http://asert.arbornetworks.com/2007/05/estonian-ddos-attacks-a-summary-to-date,
[may 09].
[18] Ruoming Pang, Vinod Yegneswaran, Paul Barford, Vern Paxson, and Larry Peterson. Characteristics
of internet background radiation. In IMC ’04: Proceedings of the 4th ACM SIGCOMM conference on
Internet measurement, pages 27–40, New York, NY, USA, 2004. ACM.
[19] Markus Kötter Georg Wicherski Paul Bächer, Thorsten Holz. Know your enemy: Tracking botnets. In
http://www.honeynet.org/papers/bots/.
[20] V. Pham, M. Dacier, G. Urvoy Keller, and T. En Najjary. The quest for multi-headed worms. In
DIMVA 2008, 5th Conference on Detection of Intrusions and Malware & Vulnerability Assessment,
Paris, France, Jul 2008.
[21] F. Pouget and M. Dacier. Honeypot-based forensics. In AusCERT2004, AusCERT Asia Pacific Informa-
tion technology Security Conference 2004, 23rd - 27th May 2004, Brisbane, Australia, 2004.
[22] The Leurre.com Project. http://www.leurrecom.org.
[23] Niels Provos. A virtual honeypot framework. In Proceedings of the 12th USENIX Security Symposium,
pages 1–14, August 2004.
[24] M. Abu Rajab, J. Zarfoss, F. Monrose, and A. Terzis. A multifaceted approach to understanding the
botnet phenomenon. In IMC ’06: Proceedings of the 6th ACM SIGCOMM conference on Internet mea-
surement, pages 41–52, New York, NY, USA, 2006. ACM.
[25] Symantec Security Response. W32.rahack.h, [april 2009].
[26] Olivier Thonnard and Marc Dacier. A framework for attack patterns’ discovery in honeynet data.
DFRWS 2008, 8th Digital Forensics Research Conference, August 11- 13, 2008, Baltimore, USA, 2008.
[27] Olivier Thonnard and Marc Dacier. Actionable knowledge discovery for threats intelligence support
using a multi-dimensional data mining methodology. In ICDM’08, 8th IEEE International Conference
on Data Mining series, December 15-19, 2008, Pisa, Italy, Dec 2008.
[28] Olivier Thonnard, Wim Mees, and Marc Dacier. Addressing the attack attribution problem using knowl-
edge discovery and multi-criteria fuzzy decision-making. In KDD’09, 15th ACM SIGKDD Conference
on Knowledge Discovery and Data Mining, Workshop on CyberSecurity and Intelligence Informatics,
Paris, France, Jun 2009.
[29] Laurens van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of Machine Learning
Research, 9:2579–2605, November 2008.
[30] Ronald R. Yager. On ordered weighted averaging aggregation operators in multicriteria decisionmaking.
IEEE Trans. Syst. Man Cybern., 18(1):183–190, 1988.
[31] Vinod Yegneswaran, Paul Barford, and Johannes Ullrich. Internet intrusions: global characteristics and
prevalence. In SIGMETRICS, pages 138–147, 2003.
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-211
Proactive Botnet Countermeasures

An Offensive Approach
Felix LEDER, Tillmann WERNER, and Peter MARTINI
Institute of Computer Science IV, University of Bonn, Germany
Abstract. Botnets, consisting of thousands of interconnected, remote-controlled

computers, pose a big threat against the Internet. We have witnessed the
involvement of such malicious infrastructures in politically motivated attacks more
than once in recent years. Classical countermeasures are mostly reactive and
conducted as part of incident response actions. This is often not sufficient. We
argue that proactive measures are necessary to mitigate the botnet threat and
demonstrate techniques based on a formalized view of botnet infrastructures.
However, while being technically feasible, such actions raise legal and ethical
questions.
Keywords. Botnets, cyberwar, DDoS, defense strategies, countermeasures
Introduction
A botnet is an alliance of interconnected computers infected with malicious software (a

bot). Bots are commanded by an operator and can typically be advised to send Spam
mails, harvest information such as license keys or banking data on compromised
machines, or launch distributed denial-of-service (DDoS) attacks against arbitrary
targets. What's more, they often interfere with regular operation rendering infected
machines unstable or unusable. Thousands of such botnets exists, with each containing
thousands to millions of infected systems. The result are major direct and indirect
consequences for economy as well as for the political life [2].
In the past, the economic damage caused by botnets has been related to bandwidth
and CPU resources bound by Spam, DDoS attacks, and propagation of the malware.
More recent reports show that the damage is largely increasing due to the number of
stolen credit card information and banking credentials [17]. As more and more botnets
are incorporating functionality to collect this data, the damage will likely increase over
the next years. Besides this, distributed denial-of-service attacks originating from
botnets disrupt business at attacked sites. The measures for handling these attacks, like
forensic analysis, moving sites into different networks, data recovery etc., cost up to
several million US dollars per incident, let alone the collateral damage, which is hard to
measure [23].
Recent developments show that botnets are not only harmful to companies and
consumers but are also involved in politically motivated activity. Largely organized
DDoS attacks conducted by botnets in 2007 and 2008 cut off major Government sites
in Eastern Europe from the rest of the Internet. This drastically shows how the vast
number of remotely controlled machines has the potential to be used as a powerful
weapon in a cyberwar rather than just being an annoying phenomenon affecting only
212 F. Leder et al. / Proactive Botnet Countermeasures – An Offensive Approach
some individuals. The spreading of botnets is conducted actively: Remote systems are
automatically attacked and exploited, mails are sent that trick the reader into opening
malicious programs or web pages which actively exploit the visiting computer. On the
contrary side, measures against botnets are very often passive and defensive. To date,
active measures are often taken in the context of responses to an ongoing incident only.
We have developed methodologies and prototypes for infiltrating botnets that can be
used to tackle them from the inside. Such offensive countermeasures can be used to
mitigate or extinguish existing botnets. We present our different approaches and
demonstrate how they can be applied to existing botnets in case studies.
The remainder of this paper is organized as follows: The next section presents a
brief overview of common botnet topologies. Section 2 reviews classical
countermeasures. Proactive approaches will be explained in section 3. In section 4
some case studies will be presented. We will discuss legal and ethical aspects in section
5. Section 6 concludes the paper.
1. Botnet Topologies
The two things needed to set up a botnet are an addressing mechanism to identify and
reach a command-and-control instance, and a communication protocol to distribute
commands to the bots. The latter is often referred to as an overlay network that forms
the botnet's communication channel. Different botnets are using different strategies
here which is reflected in the topology used: We differentiate between centralized,
decentralized and locomotive botnets. The kind of topology is extremely important for
the selection of containment strategies.
Centralized topologies as depicted in figure 1 are the classical botnet structures.
The box in the middle denotes the central C&C server with seven connected bots and a
commander (the star symbol). Examples are the IRC-based Agobot, Rbot, and Sdbot
families [1]. A static command-and control server is contacted by bots via its IP
address (which generally requires resolving a DNS name first). Centralized botnet
infrastructures often rely on existing network protocols on top of IP that implement
standard client-server architectures, like IRC or HTTP. For this reason, they are
obviously completely extinguishable by taking down their C&C server.
Figure 1. A centralized botnet with seven bots and a commander
The communication in a centralized botnet can either follow a push strategy (as in
IRC-based communication) where each bot stays connected to a server which then
distributes commands simultaneously to all hosts in a broadcast-like manner. Or the
F. Leder et al. / Proactive Botnet Countermeasures – An Offensive Approach 213
server has to be polled by the clients on a regular basis (as in HTTP-based botnets). In
the latter scenario, the general method is to set up and update a central resource like a
web page which can be browsed by the bots. Both approaches have their advantages,
e.g., IRC botnets can be built upon an existing IRC infrastructure with multiple self-
synchronizing servers, providing load-balancing and reliability. HTTP, on the other
hand, is more stealthy and better suited for bypassing security gateways and hiding
amongst regular traffic patterns.
Figure 2. A decentralized botnet with three bots acting as C&C servers
In a decentralized topology, no single command-and-control component exists.

Instead, each bot seeks for a commander using some upstream query mechanism. A
schematic structure is depicted in figure 2: Each bot knows some neighbors and
receives and forwards commands. Three bots act as C&C servers and are advised to
distribute commands in the network. Well-known representatives are the Storm Worm
[3], or Conficker [5]. The two-tiered approach allows the botnet owner to easily change
the C&C backbone, making it much harder to take it down. As in centralized botnets,
commands can be pushed to bots, which requires that they can be reached instantly, or
infected machines pull commands from their individual C&C server (the latter being
the most common case). Bots can be implemented to automatically re-establish a C&C
session on disconnects. Most decentralized botnets seen so far were based on peer-to-
peer (P2P) technology that allows for both information queries as well as host
addressing, the two features needed for the communication between a bot and a
command server. In a common P2P botnet some peers are controlled by the botnet
owner and used to issue and propagate information (i.e. commands) to other peers.
Taking advantage of the flexible self-organizing network infrastructure, these nodes are
easily replaceable with other hosts.
The decentralization can be taken even further by designing fluxy registration of
C&C servers at the query layer (i.e., a pool of command servers returned to queries
which is kept highly dynamic through automated subscriptions). This situation is
visualized in figure 3 on the next page: The shaded structures are past C&C servers that
have been replaced by other ones automatically. Bots recognize the change and contact
the new server instead. In most cases these C&C servers are also infected hosts,
temporarily playing the role of a commander.
Another way would be to change the query interface, e.g., by choosing time-
dependent domain names. We call such botnets locomotive because of their constantly
moving structure. One example is the HTTP-driven Torpig botnet [4]. Conficker, in
addition to its P2P structure, also makes use of constantly changing DNS names [5-7].
There is no standard implementation of such botnets. In fact, the overall structure is
often even more complex than outlined here.
Figure 3. A locomotive botnet with C&C servers that move over time
In reality the boundaries between centralized, decentralized, and locomotive

botnets are blurred: A similar strategy was already commonly implemented in classical
botnet infrastructures where a DNS entry was used to transparently switch between
servers. However, this does not really provide more security as it only displaces the
single point against which takeover attempts could be mounted.
2. Classical Countermeasures
Traditional ways of counter-measuring botnets is generally restricted to spotting a

central weak point in their infrastructure that can be manipulated, disrupted or blocked.
The most common way is to cooperate with an Internet service provider to gain access
and shut down the central component, resulting in a loss of control for the botnet
owner: The botnet cannot be commanded anymore. Such actions are often performed
during emergency response to an ongoing incident like a DDoS attack. While this
course of action has proven effective (e.g., shutting down an IRC-based C&C server
prevents bots from receiving commands, and machines already involved in an attack
are rebooted sooner or later), it requires access to the machine, and, most notably, the
willingness to cooperate at the responsible institution. Classical countermeasures
against botnets have three different points to attack:
1. The command and control (C&C) server

2. The botnet traffic
3. The infected computers
We will explain the countermeasures with their chances and difficulties in the
following. Our goal is to show their differences and why we need more discussion
about offensives approaches against botnets.
2.1. Taking Down the C&C Server
The most promising approach is to remove the base of a botnet, which is the C&C
server. Pulling the plug of the command-and-control host allows to extinguish the
whole botnet in one go. Unfortunately this is only possible if all of the following
conditions are met:
1. The botnet uses a centralized structure.

2. The location of the C&C server is known.
3. The provider cooperates.
If one of those conditions is not met, removing the C&C server is impossible.
More and more botnets are not solely relying on a centralized structure anymore.
Instead they use peer-to-peer (P2P) functionality or multi-proxy structures to hide their
central origin. It is often hardly possible to find the location of the C&C server of such
botnets. If multiple, fixed servers are used, all of them must be removed. When the
location is known, the provider hosting the C&C server must be cooperative. Very
often, botnets are controlled from locations hosted by so-called bullet-proof hosters,
that are not responsive to abuse requests or, even worse, move the server to affiliated
partners as soon as the pressure to take the host down rises. Those providers are found
in almost all countries, Germany and the U.S. being amongst the most prominent ones
according to our observations. Law enforcement is often one step behind when the
hosted services are suddenly switched to another provider. In the lucky situation where
the location of a centralized botnet to be within a cooperative provider's network, the
provider must still be notified and has to agree to all actions. But different
organizations that track attacks in the Internet receive so many hints about possible
C&C servers that they cannot handle, follow, and verify actions against each C&C
individually. The number of conditions is part of the problem that such a large number
of botnets exist, and it is still increasing.
Taking the C&C servers down is not always similar to removing the root of the
botnet. Infected machines can also contain functionality to spread autonomously, as
well as other fall-back logic that gets executed in case the C&C cannot be reached
anymore. This creates additional traffic and can lead to more infected machines.
Some cases are known where a botnet takeover was performed with the goal to
issue commands that make the bots stop an attack or deinstall themselves. While this
approach is more delicate with respect to responsibility for effects caused on infected
machines, it is extremely successful at the same time. Attacks are stopped immediately
and the botnet is eventually shut down conclusively without the chance to be brought
back up by the owner. However, the success rate depends on whether cooperation with
the responsible infrastructure providers is possible or not.
2.2. Sinkholing Malicious Traffic
Sinkholing is the term for redirecting network traffic or connection attempts to a

special purpose server. If the C&C server cannot be taken down, another option is to
redirect malicious traffic to sinkholes, a strategy that found its way into recent
mitigation techniques, either locally [24] or globally [14]. The sinkholes record
malicious traffic, analyze it and drop it afterwards such that it cannot reach the original
target it is meant for. One example of sinkholing is DDoS null-routing. In case traffic
belongs to an ongoing DDoS attempt is observed it is dropped and sometimes counted
for later analysis. DDoS null-routing at border-routers is a promising approach to
mitigate DDoS attacks but comes with the challenges of reliable identification of attack
related traffic and clean dissection of high-bandwidth data streams at an early stage.
This is generally only possible at ISP level. A collaborative worldwide initiative
between providers would be another option, but is obviously beyond all question.
2.3. Cleaning Infected Systems
The most sustainable countermeasure against botnets is probably to clean all

infected systems and remove the bots installed. While this removes the full power of a
botnet, it is also the most complex and most difficult to manage countermeasure. To
date, the owners or administrators are responsible for keeping their systems clean from
infections. Only recommendations and technical advice can be given to them. As most
users are not even aware of their machine being infected, let alone the ability to remove
a malware, a global cleaning is impossible. The huge media campaigns about
Conficker and the number of still infected systems show that even with intense
warnings a large-scale client-side cleanup performed by the individual owners is not
feasible.
The standard recommendation to keep systems safe from botnets is to use
firewalls and up-to-date anti-virus (AV) software. Firewalls are a preventive feature
that in many cases only block attacks from the outside. The increasing number of drive-
by-exploits, using bugs in the user's browser to infect a system, and the mobility of
malicious data on laptops or USB-sticks opens up a range of new infection vectors that
bypass firewalls. This development has been very obvious with Conficker infections
[5,14]. Anti-virus software is a reactive feature. Before it is able to detect anything,
signatures must be available and the malicious data has to be on the targeted computer.
If signatures do not yet exist, the systems cannot be defended. Tests of different AV
engines have shown that some detection rates are as low as 80% [18]. Once a system is
infected, the bot can spread and perform malicious actions until AV signatures are
available. AV engines are often outdated and not updated on a regular basis.
Furthermore, different bots disable AV scanners or hide in ways that cannot be
detected by regular scanners [19].
All in all, a global cleanup, as it would be required in order to effectively take
power from botnets, seems to be infeasible when approached at an organizational level
only.
2.4. Conclusion on Classical Strategies
The observations discussed in this section demonstrate that to date the success rate
of botnet countermeasures depends mainly on organizational and political general
conditions. Given that setting up cooperation or diplomatic agreements takes time we
come to the conclusion that establishing an appropriate relationship that legitimates
cooperation for collaborated actions is not suited as an ad-hoc scheme for fighting
ongoing attacks.
The situation gets worse considering that modern botnet infrastructures do not fall
under responsibility of one entity. Instead, distributed peer-to-peer networks operate
globally, thus shutting down local parts (often no more than single machines) would be
no effective solution. All in all, countermeasures that require close cooperation are
today generally infeasible for both technical and political reasons.
There have been discussions where experts stated that shutting down C&C servers
has become useless as they would be replaced with new, better protected systems
almost immediately. This accelerated arm's race would eventually lead to sophisticated
botnet technology sooner than without mitigation. We think that this view is by no
means bearable: It ignores the fact that botnets cause harm against other organizations.
A hands-off approach leaves potential target sites alone with the existing threat. In the
end, restricting mitigation techniques to eluding from or block ongoing attacks is an

admission of powerlessness. We propose a combination of classical techniques with
additional proactive strategies which we discuss in the following section.
3. Proactive Measures
The classical countermeasures described in the previous section are very good steps to
mitigate the power of botnets, but recent developments show that they are only
applicable to a certain extent: Newer botnets use more sophisticated obfuscation
techniques that deny the use of classical approaches due to the difficulties explained
above. While the newer structures introduced by recent botnets complicate the
applicability of some strategies, they are open to more offensive tactics. This section
explains general principles that can be exploited to create offensive countermeasures
against botnets. We focus on the technical possibilities.
Exploring the structure of a botnet is often the first step for finding starting points
for possible countermeasures. An immanent property of all botnets is that they have to
allow new machines, which run on untrusted platforms, to join the network [25]. This
is an important aspect for countermeasure approaches: We are not restricted to acting
from the outside – we can join the network, perform investigations while being part of
the infrastructure ourselves and might even be able to contain the botnet or take it down
from the inside. Furthermore, bots are spreading to infect more systems and make the
network grow. Malware samples, which are not hard to obtain, can be analyzed (i.e.,
reverse-engineered) to learn about their internals. With the knowledge about a bot's
functionality, it is often possible to create a fake bot and link it into the botnet to
monitor or perturb the internal communication. This procedure is always possible, as
all information about the initial bootstrapping has to be included in the malware binary
and can thus be cloned. Many approaches presented in this section rely on the
infiltration of botnets, a technique that was discussed in different flavors in case studies
before [3-5,9].
Offensive strategies can be split into three different categories: Mitigation,
manipulation, and exploitation. The extent to which corresponding actions are possible
depends largely on the topology used by botnet. Especially, decentralized and
locomotive topologies offer multiple chances for countermeasures.
Strategies for mitigation are offensive, technical means that slow botnets down,
by consuming resources for instance. Examples can be temporary DoS attempts against
C&C servers, trapping and holding connections from infected machines, or blocking of
malicious domains. Manipulation strategies make use of the command layer. The
knowledge about command protocols is essential to manipulate and inject commands.
The required knowledge about the protocols does also include cryptography used. Even
though cryptography may completely deny the inspection of botnet data exchange, our
Waledac case study shows how this can be achieved even when cryptographic methods
like RSA and AES are used [10,11]. Possible manipulation can be the alteration or
removal of DDoS or Spam commands as well as commands to download and execute
programs, which allows a remote cleanup of infected machine. Less invasive options
include dropping collected personal data, like credit card or banking details, replacing
them by fake information, or issuing commands to make bots stop the collection.
Lastly, exploitation is a special strategy that makes use of bugs found in bots. Like
bugs in other products, these can be used to perform actions on the infected machines.
Even though, this category is the most powerful, it is the one with the highest risk
involved because exploits can easily crash and damage systems if not designed
carefully.
Not every strategy can be applied to every botnet. Some of them depend largely
on the botnet's topology. Especially non-centralized botnets offer a range of
possibilities. We will explain different technical possibilities in the following.
3.1. The Addressing Layer
In this paragraph we discuss strategies targeting the routing and the addressing layer of
a botnet infrastructure. It is important to understand that the routing mechanism used in
a botnet is needed for addressing hosts, or C&C servers respectively. The command
layer, in contrary, works on top of the addressing scheme to provide a communication
channel to the interconnected machines.
The most common way for a bot to address a central C&C server is a DNS name
that resolves to an IP address – the addressing takes place in two phases. Each phase
makes a potential starting point for intervention. For instance, DNS requests are
generally handled by a local DNS resolver which, in turn, forwards the request to an
authoritative DNS server. This local resolver is controlled by the site administrator and
can easily be instructed to return a specially crafted response to specific queries. The
same holds for IP routing: Local routers can be equipped with routing table entries to
sinkhole certain addresses or redirect them to different hosts. As a consequence, both
steps result in bots in the local network being unable to contact the original C&C server
and might even be controlled by a pseudo-server. An intervention as described above
always requires a man-in-the-middle position. However, it is not always necessary to
change the configuration of inline devices. Approaches exist that demonstrate the live
modification of relevant network traffic [11].
Modern botnets use more complex addressing schemes which are also run as an
overlay network on top of the IP-based Internet. Examples are peer-to-peer networks
like Storm or Waledac. They provide their own addressing scheme with the goal to
increase flexibility and decentralization. Both examples will be discussed in more
detail in section 4. Again, a strategic position is necessary to infiltrate the addressing
layer of these botnets. A general approach is to inject a carefully monitored and
controlled node, e.g., a clone of an original peer.
Even when C&C servers cannot be physically accessed, they must be reachable
over the Internet because bots have to contact them to receive commands. This fact can
be used to mitigate the botnet by creating a DoS situation at the server. A controlled
allied DDoS from would make the server unreachable. Additionally, botnets often rely
on technology that is prone to specific attacks by design, like the Transmission Control
Protocol (TCP). For example, a C&C server's TCP backlog queue can be filled up with
connection attempts to trigger denial-of-service conditions, turning the botnet's
weapons against itself. This is especially useful for most HTTP based bots where new
connections are established for every command request. We have evaluated different
service and operating system combinations and found a temporary TCP DoS attack to
be easily conductible with only very few resources. During our research we have been
able to reliably decrease the probability to establish connections to TCP servers to less
than 5% with only one offensive machine. A single host can keep the victim service's
backlog queue filled, blocking all further connection attempts and thus hindering bots
from requesting commands. Such an action can be crafted in a way that it is not
possible to tell apart the connection attempts from the ones issued by bots. As a result
any counter-action intending to block the requests would also block all “legitimate”
bots. Our tests showed that one single machine can keep a TCP service permanently
unresponsive just by initiating and completing 3-way handshakes and keeping
connections up as long as possible. Such an attack results in less bots being able to
contact the C&C server and participate in malicious activity.
Flooding the link or network where the C&C server resides with packets that
consume all available bandwidth is another similar attack. It obviously requires more
resources, though, as more packets must be sent. A reflection attack can be used to
amplify the amount of traffic sent. However, that would incorporate third-party
resources and probably permission by the affected site owners which is apparently not
granted.
3.2. The Command Layer
Attacking the command layer of a botnet requires knowledge of the protocol used. An
easy example would be an IRC-based network where a remove command instructs bots
to uninstall themselves from infected systems. Many classical bots implement such an
instruction [1], and it was shown before that it can be used to disintegrate a botnet [24].
The injection of a command requires either control over the C&C server, or bots have
to be redirected to a different server by performing an attack against the addressing
layer (as described in the previous paragraph), which then distributes the removal
instruction. Other bots do not have an uninstall option but offer an update functionality
that can be used to replace the malware with a innocuous binary or a program that
scans for a bot and eventually removes it (similar to a virus scanner).

Figure 4. Interception and modification of commands from a C&C server to a bot
In combination with infiltrating the addressing layer other approaches become

feasible: Original commands can be monitored, intercepted and modified. This
situation is depicted in figure 4 where a fake bot positioned in between the path to other
bots and the C&C server intercepts the communication and replaces commands with
other information. A protocol could implement checks to render such manipulations
impossible. However, such measures were not seen in botnets so far.
In general, to actually conduct a botnet infiltration attack, a combination of actions
on both the addressing and the command layer is necessary. Redirecting bots to a
controlled server either for sinkholing or to command them to perform a self-removal is
probably one of the most effective countermeasures on the infrastructure level.
3.3. Exploitation
Exploit based strategies make use of the fact, that even botnets contain bugs and
programming flaws that result in vulnerabilities which can be exploited to gain control
either over a central component (like a C&C server) or over bot-infected machines.
Such vulnerabilities may range from misconfiguration, like e.g., an insecure IRC server
setup that allows other users to control a channel, to security holes in software, like
remotely-exploitable buffer overflows.
Mitigation and manipulation strategies are mostly not invasive for the infected
machines themselves. An exception are commands that download and execute
programs. The exploitation of bugs is even more invasive than executing regular
programs because exploit code is often required to be specifically tailored to the
targeted host operating system and language. Frameworks like metasploit [21] help in
developing generic exploit code. All in all, there still is a higher risk that remote
systems are crashed this way. This has to be taken into consideration especially in
scenarios where infected systems control critical infrastructures.
Before exploiting the bugs, infected systems have to be found. For decentralized
topologies they can be enumerated by counting connection attempts to injected bots. In
locomotive topologies this information can be extracted from sinkhole data. Other
options are the use of honeypots, IDS signatures or scanners that scan network ranges
for infected machines. In very rare cases lists of other bots are available from central
IRC C&C servers.
Exploitable vulnerabilities in bots have been found before [13]. Many Rbot and
Sdbot variants share the same code base that contains vulnerable functions like this. A
potential way to take down botnets would be to identify infected machines, exploit a
vulnerability in the bot, and inject and execute code that shuts the malware down.
Vulnerable code can still be found in recent malware. Conficker.B uses the MD6
cryptographic hash function for its digital signatures. The MD6 algorithm was found to
contain a buffer overrun vulnerability and fixed in an update release that was
immediately incorporated in Conficker.C [7]. While this particular vulnerability in
Conficker.B was not exploitable, it demonstrates that even sophisticated malware is not
immune to critical security holes. Actively attacking bot-infected machines raises lots
of ethical and legal questions. We provide a summary of the most important of these
aspects in section 5.
3.4. Conclusion on Offensive Strategies
The number of technically feasible strategies shows that there are plenty of possibilities
to pro-actively act against botnets before they cause any harm. The use cases presented
in the following section show that the offensive strategies are not purely theoretical but
based on our practical research. While technically possible, the ethical and legal
problems those strategies bring up have to be taken in consideration in practice. Before
starting to use (especially the invasive opportunities), an extensive discussion about
those topics and authorities is required. The last sections of this paper are to be seen as
a step towards this.
A general challenge about many offensive approaches is that they have to be
performed stealthy. Otherwise mitigation attempts can be countered by the botnet
commanders. Manipulation possibilities can quickly be outdated with small protocol
changes or the use of digital signatures. Furthermore, exploitable bugs can usually be
fixed in a short time. In case a botnet is to be shut down, this must be performed
globally and quickly to not leave any time to the botnet commanders for
countermeasures themselves.
Experts consider prosecution of botnet constructors unlikely to have a strong
impact on the global threat [20]. Instead, botnets must be fought on a technical level.
Proactive measures must be taken as a joint effort of international security teams with
local authority. This approach has proven successful and should be followed more
consequently in the future [22].
4. Case Studies
This section contains some brief case studies where we present our research on the
feasibility and effectiveness of proactive countermeasures on real botnets. We focused
on more sophisticated bots, rather than standard IRC or HTTP based networks, as these
are far more challenging and our methods must work for them as well. However, we
cannot discuss all the technical details for lack of space and refer to the references for
more information.
4.1. Kraken
If a botnet's communication protocol is known and messages can be forged, it is

possible to inject commands that will be reacted upon by the bots. In case of the
Kraken botnet commands are requested from a server after selecting and resolving an
entry from a list of domain names. By registering some of those domains and accepting
connections from Kraken machines it is possible to send arbitrary commands to the
bots. In [15], we have described the encryption used in the protocol. [16] have
demonstrated how a remote cleanup can be conducted by issuing an update command
that instructs bots to start a removal tool.
4.2. Storm Worm
The Storm Worm (also known as just Storm) is probably one of the most known bots
worldwide [3,9]. While other specimens that use P2P technology were seen before,
Storm was the first malware that used it in a way that the botnet could exist for more
than three years. Storm is interesting for different reasons: First, spreading was almost
only based on social engineering through sending Spam – people even started talking
about Spam campaigns as the topics were linked to current news or dates like the Iran
War or Christmas.
Storm uses an encrypted version of the Edonkey peer-to-peer protocol. We have
been able to extract the 40-bytes XOR key through reverse-engineering of a storm
sample and have built our own Storm P2P client to be able to infiltrate the network [9].
In P2P botnets like Storm, all nodes take part in the infrastructure and perform routing
or searches for other bots. Being able to communicate with other nodes, the Storm
network routing infrastructure can be infiltrated and disrupted [3,8,9]. However, we
have found a less complex yet more powerful approach [9]: We were able to extract
Storm's algorithm responsible for the privilege calculation and to displace the original
commanders. This makes it possible to issue own commands to all bots in the network.
Storm's command set has been reverse-engineered by us. Consequently, we would

have been able to instruct Storm nodes to download and run an arbitrary binary, e.g., to
remove the bot from the system. All in all, a complete take down was possible by
combining attacks on the infrastructure and the command layer while exploiting a
design flaw in the P2P protocol. Today, only an insignificant number of Storm
machines is still existing.
4.3. Waledac
Waledac is another P2P bot that tunnels all communication through HTTP.
Additionally, each message is encrypted using a hybrid encryption scheme that applies
the AES and RSA implementations of OpenSSL. To be able to spy on the traffic, we
conducted a man-in-the-middle attack and intercepted the RSA key exchange. One
important observation was that the AES key used for further encryption was static
rather than dynamically chosen, a design flaw that enables us to also decrypt Waledac
messages offline, without the need for a man-in-the-middle proxy.
Being able to snoop on the traffic, we were able to manipulate the communication
between two nodes and even developed a tool to construct and inject valid messages
ourselves. A takeover strategy based on these findings would be to announce oneself to
other Waledac hosts as a proxy node to achieve a prominent position and then drop
important commands like DDoS instructions. We could also modify update commands
to make bots download and execute our own binary instead of the one provided by the
commander.
4.4. Conficker
The first variants of the Conficker worm implement a C&C query algorithm similar to
the one used by Kraken. Every day, a list of domain names are generated. Some
randomly selected names are then resolved and the corresponding hosts are contacted.
The Conficker Working Group [15] has organized a collaborative effort for pre-
registering and sinkholing these domains to make them unavailable to the Conficker
constructors. Furthermore, vulnerabilities exist in Conficker's code that would
theoretically allow for exploitation and execution of arbitrary commands on infected
machines. We have developed a network scanner for reliable identification of
Conficker hosts [5]. These techniques can be combined in a proactive defense strategy
to take down the botnet.
5. Legal and Ethical Aspects
The technical feasibility of the presented countermeasures does not justify their use in
practice [26, 30]. The conduction of these countermeasures may interfere with law or
current ethical beliefs depending on their invasiveness and impact on third-parties. On
the one hand, many people fear the debates and political consequences, and therefore
the general tendency is to stick to conservative approaches [29]. On the other hand, the
enormous damage caused by botnets cannot be simply overlooked [23]. Since classical
means have not proven to keep up with the increasing threat, discussions have to be
initiated about more active strategies. The recently published “Cyberspace Policy
Reveiw” [26] by the White House discusses a strategy to make the Internet more secure
from an general viewpoint and identifies privacy and the question of responsibility as
important challenges. In fact, this is not really surprising as both of these topics have a
strong judicial impact.
This section aims at bringing up the most important consequences of traffic
manipulation and countermeasures against control servers. Further, we briefly discuss
some ethical, legal, and liability aspects of remote bot disinfections.
5.1. Targeting Control Servers
Most countermeasures that target C&C servers only can be regarded as non-critical.
We assume that the commanders of botnets are cyber criminals. Taking down their
C&C server literally disarms them. The same holds for regular DoS attacks on those
servers. However, DDoS attacks that use lots of bandwidth and processing power, yield
to a trade-off between the large amount of resources consumed by the botnet and
resources for DDoS countermeasures.
5.2. Targeting Traffic
Traffic manipulation is generally considered to be ethically and legally feasible as long

as affected parties agree to it. Such alterations might be offered as a service to prevent
DDoS attack, for example. Many users don't know about the threat and therefore don't
take steps towards such agreements. Inspecting their traffic and modifying it is a legal
problem in many countries, even though more and more countries, like the U.K [31].,
pass laws that allow traffic inspection from certain official organizations. Traffic
inspection and traffic modification at ISP level would allow to remove Spam and
DDoS commands as they are passed in known botnet traffic.
Such actions can also be seen as an indirect protection of the ISP's infrastructure.
However, on the one hand, such courses of action raise ethical problems as users may
interpret them as a kind of censorship [32]. On the other hand, many users don't know
about their infections and would really appreciate if their systems are not misused. A
default policy included in contracts to allow ISP to perform those actions in
conjunction with a possibility to withdraw would be a solution that is actually already
evaluated at several sites.
5.3. Targeting Infected Systems
The most controversial discussion takes place about more invasive strategies, like a
remote removal of bots from infected computers. This raises different issues:
1. Ethical: This bypasses the responsibility of users to keep their systems clean.
2. Legal: In most countries it is illegal to run software without the system
owner's permission.
3. Liability: Who takes the consequences if the cleanup actions fail or cause
problems?
A remote cleanup, in most cases, requires running a removal tool on the infected
computer, which has been shown to be technically feasible for a range of botnets. This
kind of cleanup has to be performed fast because otherwise new commands to kill the
removal tools can be issued by the botnet commanders. Thus, asking all users is not
feasible.
Up to now, users are responsible for their own systems. Remote cleanup with
automated removal tools bypasses the user, his autonomy, and his responsibility. While
some users interpret this as an intrusion into their privacy, a wide range of users would
be very grateful for this kind of support to keep their systems clean. All in all, it keeps
them a little safer from getting their banking or credit card details stolen. The typical
use of AV software as an install-and-forget means supports this view.
Downloading and running software on a remote computer without the owner's
permission is illegal in many countries because it is seen as an act of hacking into the
system [27, 28]. However, some countries, like the Netherlands, require criminally
motivated deeds for the applications of those laws. Since it is a general belief that
botnets are run by criminals and cyber terrorists, the disinfection of hosts clearly states
good will [28].
This may lead to the conclusion to run proactive strategies only for systems in
specific countries or organizations that agreed on such actions. This selective approach
is not very effective and yields at maximum in a mitigation but not removal of the
considered botnet. Technically it is not always feasible to identify hosts from specific
countries or organizations in the overlay network of the botnet. The cleanup of only
selected systems rises the problem that the left-over partition may react, adjust to the
new situation, and conduct a counter attack. Similarly, concurrent criminal
organizations may observe those actions and may use the information to issue their
own “updates”, which simply replaces the bot.
Cyber criminals act globally. Thus, countermeasure can only be effective when
performed on a global level or at least in large parts of the Internet. A global take down
would be the ideal situation. However, a global disinfection rises political questions
because most countries would not agree on another country's forces to remotely run
software on their systems. This holds especially for infected governmental systems.
The foundation of a global organization with legitimation to perform those actions
might be a solution. The discussions and consents about such an initiative have to take
place on a political level.
Even the best software contains bugs. Invasive countermeasures, like removal
tools, can lead to instability of the disinfected system, even with a low probability. This
risk increases when bugs in a malware are exploited. In the unlikely case that this
happens, the liability is an important question. Who takes responsibility for this
happening?
Closely linked to the liability question is the ethical question on the consequences
of such actions on medical devices or critical infrastructures, for instance. However, the
probability of malicious software causing harm is much more likely. During Conficker
outbreaks in hospitals, medical devices were infected and stopped working properly. In
the end, it is also a question of responsibility to leave no stone unturned – and that
might even include a proactive botnet takedown to prevent further harm.
6. Conclusion
While technically possible, we argue that pro-actively fighting botnets requires

immediate political and international consensus. It is a matter of the impact whether
people would agree to offensive approaches or not. The affected systems' criticality
have to be balanced against potential damage caused by countermeasures. This is,

however, also the case for classical mitigation techniques. The portfolio of measures
demonstrated in this paper range from more passive ones, like sinkholing, to offensive
ones, like exploiting bot hosts to take them over and clean them. We believe that a
framework for a staged approach that combines both defensive and offensive
techniques should be prepared as part of an emergency response toolkit.
We have seen that cooperation is one of the most important aspects when it comes
to successful and sustainable botnet mitigation. This holds for the technical and the
political level likewise. Trusted forums must be strengthened and extended to be
capable of reacting to botnet incidents effective and immediately. A laissez-faire policy
does not lead anywhere.
References
[1] E. Stinson, and J.C. Mitchell, Characterizing Bots' Remote Control Behavior, Springer Verlag, Proc. of
the 4th intl. conference on Detection of Intrusions and Malware, and Vulnerability Assessment, 2007
[2] S.W. Korns, J.E. Kastenberg, Georgia’s Cyber Left Hook, 2009
[3] T. Holz et al., Measurements and Mitigation of Peer-to-Peer-based Botnets:
A Case Study on Storm Worm, Proc. of the 1st Usenix LEET’08
[4] B. Stone-Gross et al., Your Botnet is My Botnet: Analysis of a Botnet Takeover, UCSB Tech. Rep., 2009
[5] F. Leder, T. Werner, Know Your Enemy: Containing Conficker, Honeynet Project, 2009
[6] P. Porras, H. Saidi, and V. Yegneswaran, An analysis of Conficker's Logic and Rendezvous Points, SRI
International Technical Report, 2009
[7] P. Porras, H. Saidi, and V. Yegneswaran, Conficker C Analysis, SRI International Technical Report, 2009
[8] C. Kreibich et al., On the Spam Campaign Trail, First USENIX LEET’08
[9] G. Wicherski et al., Stormfucker: Owning the Storm Botnet, 25th Chaos Communication Congress, 2008
[10] F. Leder, Waledac is wishing merry christmas, http://www.honeynet.org/node/325, 2009
[11] F. Leder, Speaking Waledac, https://www.honeynet.org/node/348, 2009
[12] Team Cymru, A Taste of HTTP Botnets, 2008
[13] Sasser Ftpd Exploit, http://www.securiteam.com/exploits/5AP0J0ACUM.html
[14] The Conficker Working Group, http://confickerworkinggroup.com
[15] F. Leder, P. Martini, NGBPA Next Generation BotNet Protocol Analysis, IFIP SEC 2009
[16] P. Amini, C. Pierce, Kraken Botnet Infiltration, http://dvlabs.tippingpoint.com (2009)
[17] Symantec, Symantec Global Internet Security Threat Report 2008, 2009
[18] Malware Research Group, http://malwareresearchgroup.com (2009)
[19] J. Rutkowska, Subverting the Vista Kernel for Fun and Profit, Blackhat Briefings 2006
[20] R. Lemos, “Arrests unlikely to impact bot net threat, say experts”,
http://www.securityfocus.com/news/11344 (2009)
[21] The Metasploit Project, http://metasploit.com
[22] J. Stewart, Interview: “Researcher argues for CERTs with teeth”,
http://www.securityfocus.com/brief/950
[23] Ponemon Institute, 2008 Annual Study: Cost of a Data Breach, 2009
[24] V, Thomas, N. Jyoti, “Bot Countermeasures”, Journal in Computer Virology, 2007
[25] R Vogt, J Aycock, M Jacobson, “Army of botnets”, ISOC Symposium on Network and Distributed
Systems Security, 2007
[26] Panel: Ethics in Botnet Research, LEET 09, Boston, April 21, 2009
[27] T. O'Connor, Cybercrime, Cyberlaw, and Cybercriminals,
http://www.apsu.edu/oconnort/3100/3100lect02b.htm , Dec. 2007
[28] B. Koops, "Cybercrime Legislation in the Netherlands", Cybercrime and Security, Vol. 2005/4, D.Ferry
[29] D. Fisher, Botnet disruption raises ethical concerns among researchers,
http://searchsecurity.techtarget.com/news/article/0,289142,sid14_gci1311711,00.html, Apr. 2008
[30] vnunet, Security experts blast BBC over botnet stunt,
http://www.vnunet.com/vnunet/news/2238682/experts-blast-bbc-botnet-stunt, Mar. 2009
[31] EPIC, European Commission Seeks to Protect Internet Privacy, Apr. 2009
[32] CCC, Internet censorship: blocking objectionable content only protects felons,
http://www.ccc.de/press/releases/2009/20090212/?language=en Feb. 2009
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-226
When Not to Pull the Plug – The Need for

Network Counter-Surveillance Operations
Scott KNIGHT1,a, Sylvain LEBLANCa
a
Royal Military College of Canada
Abstract. The classic response to attack in computer networks has been to

disconnect the effected system from the network, preserve the information on the
system (including evidence of the attack for a forensic investigation), and restore
the system. However, it can be argued that this type of response is not appropriate
in many situations. This paper argues that understanding the adversary is essential
to effective defence. Instead it may be appropriate to respond with a Network
Counter-Surveillance Operation to observe the activity of the attacker. The aim of
this research is to enable this new kind of operation through the identification and
development of the new tools and techniques required to carry it out. This paper is
an omnibus presentation of a group of research projects associated with satisfying
this aim, namely tools to help observe the attacker's actions on the compromised
system, tools to provide a realistic environment on the compromised system, and
tools to mitigate the risks associated with the attacker's use of the compromised
system. The argument for the tools and techniques described is presented in the
context of an illustrative Network Counter-Surveillance Operation.
Keywords. Network Counter-Surveillance Operation, Computer Network

Security, Network Defence
Introduction
The classic response to the compromise of a computer system has been to remove the
system from service (perhaps preserving memory images of the system for forensic
analysis), clean/reimage the system, and restore the system to service (perhaps after
patching the suspected vulnerability) [1]. However, we likely do not know who the
attacker really was, what his mission was in attacking this computer system, what are
the attacker’s capabilities, to what extent has the attacker compromised our systems,
what are the attacker’s strategic goals, etc. When under targeted attack from a
dangerous adversary it may not be appropriate to pull-the-plug as a response;
maintaining contact with the attacker and managing the risk of the attacker’s presence
on the network may be an opportunity too valuable to ignore. There is a need for new
tools and techniques to adequately observe the attacker and to manage the risk of such
operations.
1
Corresponding Author: Scott Knight, Department of Electrical and Computer Engineering Royal Military,
College of Canada; Email: knight-s@rmc.ca
S. Knight and S. Leblanc / When Not to Pull the Plug 227
The world-wide computer network environment is recognized as being a new

battlespace. As such it would seem logical to apply lessons from more traditional
battlespaces where they seem suitable. For example, in no traditional battlespace would
a defender apply the defensive tactics, as described in the last paragraph, as a primary
response. That is, building firewalls, hard perimeter defences, defence in depth, and
then break contact with the enemy as quickly as possible as soon as he shows up. The
continued use of this kind of a response may accomplish short term tactical aims such
as restoring network services, but abandons any thought to identifying or achieving
strategic goals geared towards discovering the attacker's identity, capabilities or
objectives. As in other battlespaces, it is the achievement of strategic goals that will
win the conflict.
A basic tenet of area defence as prescribed by the U.S. Army Field Manual is that
gaining and maintaining contact with the enemy is vital to the success of defensive
operations [2]. “As the enemy's attack begins, the defending unit's first concerns are to
identify committed enemy units' positions and capabilities, determine the enemy's
intent and direction of attack, and gain time to react.” [2] In the sphere of naval
operations where conflict at sea can be a cat and mouse game of detecting, stalking and
engaging, commanders have been censured for failing to maintain contact with the
enemy [3].
The recurring directive to maintain contact with the enemy arises from the need to
know who the enemy is, what his capabilities are, and what his intention is. This is
especially important when the enemy is hard to detect in the first place and there is an
incomplete understanding of his capabilities and objectives.
However, maintaining contact with an attacker is a hard thing to do in a modern
computer network environment. The attacker will be attempting to conceal himself
using encryption, hidden processes and rootkits [4]. If the attacker becomes aware that
he has been detected he is likely to change his tactics, techniques, and procedures
(TTPs) resulting in defenders loosing contact or being fed misinformation. Of course
there are also risks inherent with maintaining contact with the attacker. In the scenarios
of interest to this research the attacker has been detected on our system(s) by the
defenders. It may be difficult to contain the attacker on the compromised system(s) and
mitigate the hazards such a presence poses to the rest of the network. Indeed it may not
be possible to maintain contact without accepting some residual risk. However, the risk
posed by maintaining this contact may be acceptable when considering the alternative
risk of breaking contact, and what that implies.
The aim of this paper is to identify the need to make Network Counter-surveillance
Operations (NCSOs) an accepted and standard response in certain cases of computer
network compromise. The paper also aims to present the research and development of
new tools to enable such Network Counter-Surveillance Operations.
Section 1 of the paper makes the argument for why NCSOs are needed and why
they are an appropriate response to some instances of network attack. The section will
also briefly identify what capabilities and tools are implied by the need for NCSOs.
Section 2 will present three current projects that address the need for these new
capabilities. These projects present techniques for covertly observing an attacker,
maintaining a cover-story or outward appearance of normal activity while an NCSO is
in progress, and a tool for containing the attacker and mitigating the risk associated
with the presence of the attacker on the network. The last section provides a conclusion
to the paper.
228 S. Knight and S. Leblanc / When Not to Pull the Plug
1. The Need for NCSOs
The remove-clean-restore (RCR) response to network attack may be a useful tactic in

mitigating the risk associated with a broad non-targeted attack, such as a rapidly
propagating virus or worm, or a script-based attack that is exploiting a published
software vulnerability. But such a response is not likely to be effective in mitigating the
risk associated with targeted attack. Targeted attacks by criminal organizations, non-
state actors, or foreign governments are the most serious threat to government/military
systems in terms of loss or damage to information assets. The RCR approach leads to
some feeling of security in winning the short-term battle, but frustrates the strategic
objective of winning the cyber war with the enemy mounting the targeted attack.
By limiting themselves to the RCR responses the defenders may win every battle
(i.e. remove the attacker from the compromised systems), but still not prevent the
attacker from achieving his strategic goals. To win the cyber war at the strategic level
will ultimately require identifying and understanding the enemy. The immediate
application of an RCR response denies the defender understanding of who the attacker
is, what capabilities the attacker has, and what his objectives are (both his immediate
tactical goal, and ultimately his strategic objectives). Controlled surveillance of the
attacker’s activities, TTPs and unfolding his communications links can provide the
defenders with intelligence on the attacker and understanding of his objectives.
Modern government/military computer systems and networks are extremely large
and complex systems. The technology and topology of the systems mean that they
inherently have large and poorly defined perimeters. Weaknesses are routinely
exploited by attackers in every layer of a system’s architecture from the network
switching equipment to the desktop applications. Current protection technologies make
it impossible to prevent successful attack on such large network perimeters. This is an
asymmetric conflict environment where a relatively small, covert attacker can
effectively engage a strong, well-resourced defender. The RCR response is actually
counterproductive in this situation because it will move the attacker away from an
attack-lane that is observable (and perhaps controllable) thus breaking contact with the
enemy. Moving the attacker from the attack-lane where he has been discovered does
not effectively deny access to the system. In a targeted attack scenario the attacker will
very likely be back, using another attack-lane. In this battlespace the enemy is hard to
find; therefore the defender may not detect the new attack and thus loose the
opportunity to observe or control the attacker. Additionally, the RCR response is likely
to alert the attacker that he has been detected and will quite likely force him to change
his TTPs as a result.
In many cases the RCR response is not available to the defenders of the network
because the system(s) compromised cannot be removed from service. This may be
because the system is providing some critical service that cannot be disrupted. In this
case both the attacker and the defenders are sharing a common infrastructure to support
their missions, which are the resources and services of the compromised system. The
defenders will have to contain the attack and battle for control of the live compromised
system. Preparation for that battle will require proper surveillance and understanding of
the attacker, either on the compromised system itself or further back along the
attacker’s communications chain.
1.1. Communality with Modern Warfare Doctrine
Consider that the scenario we are investigating has a number of common elements with
the urban warfare battlespace. Characteristics from the urban warfare battlespace [5][6]
that are common with the computer network battlespace are listed below:
• Complex battlespace terrain (i.e. many complex layers of intersupporting
technologies, communications mechanisms and applications).
• This complex terrain is inhabited by non-combatants (i.e. legitimate users of
the system, their processes and data).
• An infrastructure upon which both the attacker and the non-combatants
depend to accomplish their goals, missions.
• Many internal vital points that cannot be completely defended (i.e. complex
ill-defined perimeter to the battlespace).
• Asymmetric threat agents.
• An enemy that is hard to locate and identify.
• An enemy that is hard to separate from non-combatants.
All NATO nations train their forces in general to operate adopting the manoeuvrist
approach in their plans to defeat the enemy. This approach has been adapted for urban
area operations [6]. The Understand, Shape, Engage, Consolidate and Transition
(USECT) framework is used to conduct such operations [5][6]. The manoeuvrist
approach moves the focus from the traditionally predominant Engagement element to
the Understand element (usEct to Usect). This fits well with our argument that
immediate engagement using an RCR response may not be appropriate, and that there
are cases where we want to remain in contact to conduct a surveillance operation in
order to develop understanding of the attacker, and to control the actions (i.e. shape the
battlespace).
Tables 1 and 2 present elements from the NATO doctrinal recommendations for
understanding and shaping that seem especially applicable (edited to reflect the
computer network battlespace) [6].
Table 1. Understand Capabilities

NUMBER CAPABILITY REQUIREMENT
U5 Establish a psycho-sociological profile of the potential enemy
U6 Determine intent, aim, location, movement, status, capabilities, support structure of the
potential enemy
U7 Acquire an accurate understanding of the infrastructure, the systems and the dynamics
of the computer network environment and their impact on operations (identify the key
components/technologies and their vulnerabilities)
Table 2. Shaping Capabilities

NUMBER CAPABILITY REQUIREMENT
S2 Selective control of infrastructure, utilities and communications
S4 Restrict enemy movement/intentions
S6 Provide own users/assets with adequate protection against the entire threat
S8 Isolate the computer network battlespace
S14 Deny the enemy from operating effective C4ISTAR systems
S15 Deceive enemy as to own force intentions and actions
The concept of NCSOs as a response to network attack is motivated by achieving

these capabilities though operations that emphasize maintaining contact with the
attacker. As with other operations, surveillance combined with stealth is often
sufficient to maintain contact, and is the preferred method for doing so [2]. The NCSO
is designed to provide an understanding of the attacker and shaping of the battlespace.
Shaping the battlespace through isolation is aimed at denying the attacker any
advantages of occupying the compromised computer system. Isolation will also protect
friendly users and assets within the limits of an acceptable risk envelope for the
operation.
1.2. Requirements for NCSO Toolset
Application of a manoeuvrist approach to computer network defence using the USERT

framework implies that NCSOs must be conducted with a view to enable understanding
of the attacker and shaping of the network battlespace. This in turn implies the need to
satisfy the capabilities identified in the paragraphs above. There are currently no
technologies or supporting tools to satisfy these required capabilities. An initial set of
required capabilities might be broken down into the following areas for further research
and development:
• A toolset for covertly monitoring an attacker’s processes and communications
activity on a compromised computer system (i.e. the attacker cannot be aware
of the surveillance),
• A toolset for maintaining an adequate cover-story on the compromised
computer system (i.e. synthetic user activity that maintains the appearance that
a system is still being used in a normal way), and
• An internal network firewall to isolate the attacker’s activity in order to
contain the attack and the risk to other friendly assets while maintaining the
covert nature of the surveillance (i.e. through blocking, spoofing, modifying
the attackers interaction with friendly systems).
2. NCSO Capability Development
The toolset requirements that we have discussed above represent new areas of research
that have not been addressed in the research literature. The Computer Security
Laboratory (CSL) of the Royal Military College of Canada has begun a research thrust
entitled Network Intelligence Surveillance Toolset (NIST) which begins exploratory

research into such tools. The following sub-sections will describe an operating scenario
that provides context for the research and three of the CSL's NIST research projects
that would support NCSOs.
2.1. Representative Scenario
To help motivate the circumstances in which NIST tools are expected to operate
consider a scenario similar to the GhostNet cyber spying operation discovered in March
of 2009 [7]. For our purposes, let us imagine that a sophisticated attacker, representing
a hostile government, has exploited the computer system of a senior embassy staff
officer located in an enclave inside the hostile country that is being protected by the
defender.
The attacker has many options for effecting the initial compromise, from a social
engineering attack to many different compromises. These are of little interest to us
here, because the short duration of the initial attack makes it unlikely that the defender
will discover the compromise. However, once the system has been compromised, the
attacker will ensure that he will be able to regain access through the network by
installing back doors. The attacker will install malicious tools such as rootkits [4]
allowing him to hide the processes that he is running on the compromised system, thus
decreasing the likelihood that his compromise will be detected. To further disguise his
actions, the attacker will encrypt all communications with the compromised system.
Finally, having gained a foothold in the network, the attacker will consider both the
value of the information of the system he has compromised, and the potential of using
it as a launch point for further attacks. The compromised system could be used against
other systems inside the protected enclave or against other networks, thus providing the
attacker a level of deniability.
2.2. Covertly Monitoring an Attacker
The first of the capabilities required above was to the development of tools that allow a
defender to covertly observe an attacker’s actions on a compromised computer, and to
remain unobserved by the attacker. For the sake of minimizing detection in cases such
as our representative scenario (Section 2.1), we must assume that the attacker has
managed to gain administrator (root) access, therefore making any actions made by us
on the compromised machine visible to him. This lack of effective isolation motivates
research into protecting observation tools by raising the compromised operating system
(OS) into a virtual machine and moving the observation tools into an underlying virtual
machine monitor (VMM), as depicted in Figure 1. The interface between the OS’s
virtual machine and the VMM provides strong security guarantees, but this comes at
the expense of operating with out the abstractions provided by the OS and so severely
degrades the quality of information that can be collected about the attacker. This
tension between isolation and abstraction visibility can be partially resolved with
virtual machine introspection — the process of reconstructing the OSs abstractions in
the VMM. Previous techniques for virtual machine introspection have had
shortcomings that render them unsuitable for the observation of attackers: they either
rely on software agents in the virtual machine under observation, defeating the purpose
of migrating to the virtual machine monitor, or rely on prior extensive knowledge of
the OS kernel’s internal layout and implementation, making them unsuitable for
observing closed-source OSs. Recently developed techniques [8] significantly narrow

the gap between the semantic visibility possible from tools behind the virtual machine
interface and those naturally available in the compromised guest OSs. Implementations
of these techniques are shown to be eective in observing guest system calls and
retrieving information exposed by guest OS system call interfaces.
Figure. 1. The architecture for the intrusion surveillance toolset suggested by the scenario and problem-
space.
In order to maintain the appearance to the attacker that the machine is still being
used by a normal user (cover story in Section 2.2), we cannot physically work on the
infected machine; we must work remotely and establish a session with a local
surveillance toolset. This session must remain invisible to the attacker, so obviously it
cannot run on the compromised machine either. At the same time, this toolset must
provide us the ability to inspect the state of the machine and to monitor the attacker’s
actions. This scenario introduces several implicit constraints to the problem. Based on
this, we have identified that our surveillance system must contain four major
components [9]: a remote toolset, a local toolset, a VMM module, and a covert
communication channel between the remote and local toolsets. This high-level
candidate architecture is depicted in Figure 1.
To start an investigation in the context of the representative scenario (Section 2.1)
used in this paper, we will conduct surveillance in an attempt to identify the origin of
the attacker’s network traffic. Effective surveillance will require the ability to perform
a number of tasks. For our purposes we define the surveillance system as the
components which allow the analyst to perform introspection on the compromised
machine from a remote location. In this case, introspection refers to the ability to
intercept and reconstruct system calls, and to inject system calls and gather their
results. The surveillance system must provide the ability to list all running processes,
despite that attacker’s attempts to hide them. It must do this using multiple techniques,
in case the attacker has managed to hide himself from one method. The surveillance
system must provide the ability to inspect the network activity, including open
connections and the process-to-port binding map. Since we can see the encrypted
traffic, we should know which port our attacker is using, so this may point us to a
process id (pid) for the attacker’s process. The surveillance system must provide the
ability to browse the infected machine’s file-system in order to look for files belonging
to the attacker. The ability to do file inspection “on-the-fly” is also required. It is also a
requirement to be able to copy files back to our trusted domain in order to perform a
thorough inspection. The surveillance system must provide the ability to inspect the
registry of the compromised machine. After discovering where the attacker “lives” on
the system, the analyst needs to monitor him. In our scenario, the attacker is controlling
the machine via encrypted communications with a backdoor process. To monitor his
actions will therefore require observing the communications between encryption
process and the backdoor process. The surveillance system should provide the analyst
the ability to watch the attacker’s actions in “real-time.” The surveillance system must
provide this same observation capability in a ‘background’ mode, writing data to a file
while providing the analyst the ability to perform other tasks in the foreground. The
identification and development of the surveillance system is the subject of current
research [9].
The current research project has developed and validated useful techniques for
virtual machine introspection and is now developing a more complete suite of counter-
surveillance tools for the intelligence analyst monitoring the attacker.
2.3. Maintaining a Cover-story
A targeted attack such as the one we described in the representative scenario is likely
carried out by a sophisticated attacker. Such an attacker is likely to go to great lengths
to ensure that the system that he has compromised is indeed a high value target. In fact,
the literature has many examples of techniques used by attackers to detect traps like
honeypots: such as looking for evidence of tampering with the system call table [10], or
finding differences between the memory reported by a kernel and the memory it
actually uses [11].
We argue that an attacker would also be expecting to see activity at the human
interface devices if the compromised system is a user workstation. Such activity at the
human interface devices can also be used to characterize the compromised system and
we use the term Vitality Detection to describe such characterization efforts by the
attacker [12]. We posit that the attacker can gather statistics on the mouse and keyboard
events being generated to derive user activity on the compromised system. The attacker
can then compare this derived activity to models of user behaviour to decide if it
appears anomalous.
The targeted attack described in the scenario would be a significant undertaking
for the attacker, and he would likely wish to maximize the benefits that he can derive
from access to the compromised system. Just as the defenders wish to be able to
covertly monitor the attacker on the compromised system (Section 2.1), the attacker
also wishes to remain unobserved from the user of the compromised system. This
limits the attacker's vitality detection options. For example, the attacker will not risk the
bandwidth required to stream the entire human interface event stream and he will limit
the processing carried out on the compromised system.
To fool the attacker's vitality detection capability, we suggest a Synthetic User

Environment (SUE) that would generate human interface device events in a manner that
is consistent with a human user. Given a target document, SUE would be able to create
a document production model that would cause the release of the stream of human
interface device events that would make it appear as if that target document had been
input on the compromised system by a human user. Because humans do not input a
document starting at the first capitalized letter of the title and ending the final period,
SUE has to be able to generate errors and text editing choices that are consistent with a
human user. This process is depicted at Figure 2.
Syntactic Document
Element Editing Actions
Lexicon Lexicon
Document
Target Syntactic
Production Editing Actions
Document Element
Model Selection
Extraction
(DPM)
DPM
HID Events Keyboard &

Stream Mouse HID
User Personality Generation Event Lexicon
Model
DPM
Event Timing Error

DPM DPM
Characterisation Introduction
Figure 2. Generation of Human Interface Device Events by a Synthetic User Environment
In addition to helping maintain a cover story on the compromised system, this

research is expected to provide additional benefits. It will contribute better models of
user activity at the level of human interface devices, which are poorly documented in
the open literature. This research would also make it easier to efficiently monitor the
attacker's activity on the compromised system. This is because the activity generated by
SUE is known to the defender, therefore making it easier to attribute observed activity
on the compromised system to the attacker.
2.4. Isolating and Containing the Attacker
The CSL research project dealing with attacker containment and isolation is called
ApateX. ApateX is an intelligent transparent network bridge which controls
communications traversing it. Its key capabilities are to allow, block, spoof and/or
modify communications traversing it [13]. The tool is designed to isolate the attacker’s
activity in order to contain an attack and manage the risk to other friendly assets while
maintaining the covert nature of the surveillance.
An essential concept for this tool is that of a risk domain. Risk can be defined as a
function of an asset’s value, the agents threatening the asset and its vulnerability. For
the purpose of this tool, a risk domain is defined as a subset of networked components
(ex computers, storage devices, network infrastructure, etc) that share similar asset
value, threats and vulnerabilities. A risk policy for an operation can be enunciated by
specifying the kinds of traffic that can be allowed between the risk domain that
contains the system compromised by the attacker and all other risk domains. Note that
the internet external to the protected enclave (i.e. the outside world) may be defined as
a risk domain.
Similarly to a firewall ApateX operates by examining the packets that traverse it
and making decisions based on characteristics of those packets. The criteria upon
which ApateX makes its decisions can be one or more of: the IP addresses, port
numbers, and payload strings of the packets. However unlike a classic firewall, ApateX
can modify and redirect packets in addition to simply passing or blocking them.
We can examine the capability of the ApateX tool to isolate and control an attacker
on a compromised computer system, while maintaining the covert nature of the
surveillance. This capability allows the defenders to mitigate the risk associated with
maintaining contact with the attacker. Consider the following cases in the context of
Figure 3.
a) We may want to allow packets associated with the attacker’s communications
links to the outside world to pass. In this case we would allow packets to the
attacker’s IP address to pass to Risk Domain Foxtrot. To cut off the
communications links would isolate the compromised machine, thereby breaking
contact with the attacker and exposing our knowledge of the compromise.
b) Risk Domain Charlie may contain very sensitive information assets and we may
not want the attacker to gain any access to that sub-network. In this case we
would block any attacker packets to or from Risk Domain Charlie. This will
cause that network to effectively disappear from the perspective of the
compromised machine. Alternatively, ApateX can redirect traffic for Risk
Domain Charlie to a dummy system/network and provide cover for the operation.
c) We may want to allow domain name services (DNS) for the attacker. In this case
we would pass packets to and from Risk Domain Echo. DNS is common
infrastructure and is needed by the attacker and the defender. To disallow DNS
traffic would be unusual and alert the attacker that he may have been discovered.
We can limit the attacker’s access to the DNS port and restrict connection
attempts to other services/ports to contain his ability to attack the DNS server.
d) Access to the file server and mail server in Risk Domain Delta may be
considered too dangerous to allow the attacker to have access. The attacker may
have captured or cracked the passwords on the machine he has compromised.
These passwords may be valid on the file and/or mail server accounts. We can
inspect packet on the fly using ApateX to look for login attempts. The
account/password information can be modified on-the-fly as it passes through

ApateX to Risk Domain Delta. This will result in invalid login attempt messages
being returned to the attacker. This preserves the covert nature of the surveillance
operation because, from the attacker’s perspective, it is not possible to tell that
the login information has been modified. From his perspective it may just appear
that the login information he has gathered is not valid on the machines he is
trying to use it on.
e) The attacker may try to attack another external computer somewhere on the
internet from our network, e.g. a DDoS attack. We can use ApateX to limit the
speed or number of packets allowed to Risk Domain Foxtrot. The attacker’s
DDoS software seems to work normally from his perspective but never gets to
the target. The attacker cannot readily tell at what point in the communications
path any filtering of the attack traffic is being done.
ApateX has been fully implemented and is a working system with the capability to
allow, block or modify packets traversing it based on criteria passed to it through a user
defined policy[13]. Work is continuing with ApateX to develop more protocol
awareness through deep-packet inspection.
Figure 3. Representative ApateX Deployment
Conclusion
There are many parallels between the computer network battlespace and other modern
warfare environments, and we should draw lessons from those other aspects of military
operations, including urban area warfare and manoeuvrist doctrine. We have
demonstrated that there are situations where it is not appropriate to follow the remove-
clean-restore response to a network compromise because it breaks contact with the
attacker. In such situations, it may be appropriate to mount a Network Counter-

Surveillance Operation to gather intelligence on the attacker. An NCSO can help the
defenders gather intelligence on the attacker, including his identity, his immediate
tactical goal and his strategic objective. This maintenance of contact can also allow the
defenders to follow the attacker’s communication paths, which may ultimately lead to
the discovery of other compromised systems.
The RMC CSL research projects that we have introduced in this paper begin the
exploration necessary to mount successful NCSOs. While we have made inroads in the
areas of covert attacker observation, maintenance of a realistic environment for the
attacker, and attacker isolation and containment, much work remains to be done. In
order to properly gather intelligence on the attacker we will need to reorganize our
computer network defence organizations, to earmark resources and develop proper
operational templates. NCSOs will have training implications for operators and require
the development of new TTPs and doctrine. NCSOs will be expensive and hazardous,
but we cannot afford to ignore this requirement; doing would prove more expensive
and hazardous in the long run.
References
[1] Computer Emergency Readiness Team (CERT), Steps for Recovering from a UNIX or NT System
Compromise, Online Available: http://www.cert.org/tech_tips/win-UNIX-system_compromise.html, 20
May 2009.
[2] The United States Army, Army Field Manual FM 3-90, Department of the Army, Washington, DC , 4
July 2001.
[3] Sweetman, Jack, The great admirals: command at sea, 1587-1945, Naval Institute Press, 1997.
[4] Hoglund, Greg, and Butler, James, Rootkits: Subverting the Windows Kernel, Addison-Wesley, 2006.
[5] U.S. Department of Defence, Joint Publication 3-06 Doctrine for Joint Urban Operations, DoD, 16 Sep
2002.
[6] North Atlantic Treaty Organisation Research and Technology Organisation, RTO Technical Report
71Urban Operations in the Year 2020 - RTO-TR-071, NATO, April 2003.
[7] JR02-2009, Tracking GhostNet: Investigating a Cyber Espionage Network, Information Warfare
Monitor, 29 March 2009 [Online] Available: http://www.f-secure.com/weblog/archives/ghostnet.pdf
[8] Major, Daniel, Exploiting System Call Interfaces to Observe Attackers in Virtual Machines, Master’s
Thesis, Royal Military College of Canada, 2008.
[9] Heywood, Harley, Investigating Architectural Design Pressures on a Virtual Machine Based Intrusion
Surveillance Toolset, ECE Dept., Royal Military College of Canada, 2009.
[10] M. Dornseif, T. Holz, and C. N. Klein, "Nosebreak-attacking honeynets," Arxiv preprint
cs.CR/0406052, 2004.
[11] C. K. Tan, \Detecting sebek win32 client," SIG 2 G-TEC -, Jun. 2004. [Online]. Available:
http://www.security.org.sg/vuln/sebek215.html
[12] Leblanc, Sylvain Paul, Toward the Creation of a Synthetic User Environment – An Active Network
Defence Enabler, Depth Research and Doctoral Proposal, ECE Dept., Royal Military College of
Canada, January 2008.
[13] Vessey, David and Smith, Pat, DID-08 Detailed Design Document - ApateX, ECE Dept., Royal
Military College of Canada, 2008.
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-238
Autonomic Computer Network Defence

Using Risk State and Reinforcement
Learning
Luc BEAUDOINa, Nathalie JAPKOWICZb and Stan MATWINb
a
Defense Research and Development Canada
b
University of Ottawa
Abstract. Computer Network Defence is concerned with the active protection of

information technology infrastructure against malicious and accidental incidents.
Given the growing complexity of IT systems and the speed at which automated
attacks can be launched, implementing timely and efficient network incident
mitigating actions, whether proactive or reactive, is a great challenge. We refer to
the automation of action selection and implementation in this domain as
Autonomic Computer Network Defence. In this work, we suggest that Autonomic
Computer Network Defence can be achieved using Reinforcement Learning and
dynamic risk assessment to learn the optimal action sequence, or policy, to recover
from given computer network risk situations. Such a policy could then be used by
commercial network management and security products to implement selected
mitigating actions automatically, as risk states are sensed.
Keywords. Autonomic Computer Network Defence, Reinforcement Learning,

risk, simulation.
Introduction
Autonomic Computer Network Defence (CND) aims to provide a self-protection

capability of information technology (IT) networks in order to limit the risk caused by
malicious and accidental events. This requires an automated controller with a policy,
which selects the most appropriate action in any undesired network state. Due to the
complexity and constant evolution of the CND environment, a-priori design for an
automated controller is not effective. A solution for generating and continuously
improving CND decision policies is needed.
A system capable of achieving autonomic CND must be able to iterate through the
CND decision cycle in an automated manner. This cycle typically involves the
following steps: sensing network changes, analyzing their impact, selecting an
appropriate mitigation action, and implementing this action back onto the network.
This process forms a control loop which employs available resources to continuously
protect the IT infrastructure. Various commercial products and research prototypes
support individual steps of this control loop. However, the search for an adaptive
controller design capable of steering IT networks towards an acceptable and stable
equilibrium in the face of security events is in its infancy. Related research areas
L. Beaudoin et al. / Autonomic CND Using Risk State and Reinforcement Learning 239
include autonomic computing, autonomic networking and automated security policy

management [1] [2].
In our work, we investigated the suitability of different Reinforcement Learning
algorithms paired with dynamic risk assessment to form the basis of an autonomic
CND controller. We developed an experimental framework, which includes Discrete
Event Dynamics System (DEDS) simulation and graph models of the environment, to
iterate through CND policies in various scenarios and attempt to minimize business
risk. We show that Reinforcement Learning algorithms can learn efficient CND
policies. We also show that the difference between policies becomes less significant as
the resources available to implement CND actions are increased.
1. Risk in CND Decision Making: The Proactive-Reactive Dilemma
The North Atlantic Treaty Organization (NATO) has defined Computer Network
Defence as: “Actions taken through the use of computer networks to protect, monitor,
analyze, detect and respond to unauthorized activity within information systems and
computer networks”1. Generally, CND actions can be implemented either proactively
or reactively, and are triggered by metrics such as up/down asset status, security alerts,
disclosure of new vulnerabilities, or capacity engineering. An important difference
between proactive and reactive actions is the notion of risk. A proactive action aims to
reduce the probability p of occurrence of a damaging event (p<1), whereas a reactive
action typically aims to reduce the damage of an actual event (p=1). Often, there are
situations of conflicting priorities, where resources must be rationed between proactive
and reactive actions. An example of such proactive-reactive dilemma would be
Network Operations Centre Staff (NOC Staff) needing to decide between first patching
a vulnerable system, or fixing a simultaneous outage on another system. To solve this
proactive-reactive dilemma, we need a single risk metric which can account for both
potential (p<1) and actual (p=1) incidents. This risk metric would also need to account
for the combined effects of time, new security event arrivals and the mitigation actions
selected.
This leads to a combinatorial problem whereby risk for every network state, action
and event timing is path-dependent. This means that the overall risk exposure of a
given situation depends on the entire sequence of actions taken to recover from it.
To illustrate this important concept, a decision tree for a simple scenario with three
assets, only one of which being exposed to the Internet, is shown in Figure 1. The first
decision is triggered by a new vulnerability on asset 1, which is exposed to external
exploit sources. From the resulting decision tree, we show a sample of seven potential
risk outcomes, depending on the timing of the first exploit event, the duration of each
action, and the mitigation strategy used. Deciding to immediately patch asset 1’s
vulnerability (node 1) may lead to an optimal path (green arrow starting at node 2,
which means that low risk is assumed from this point on). However, the same decision
may also lead to the worst path, in the case an exploit occurs before the patching action
is completed, and spreads to the two other assets. For this reason, isolating the
vulnerable host first, prior to patching it, may be an advantageous decision path.
Although isolating asset 1 results in self-denial of service, it also prevents any exploits
1
NATO publication 3000 TI-3/TT-1162
240 L. Beaudoin et al. / Autonomic CND Using Risk State and Reinforcement Learning
from using this host to spread to other hosts. Once exploited, an asset has to be “fixed”,
which typically means restoring a clean disk image on the host.
Figure 1. Example of a simple CND decision tree for a new vulnerability event
In more complex scenarios, new vulnerabilities, exploits and outages can occur at
any point in time, resulting in new decision branches for each arrival, many of which
can be followed concurrently, or serviced through a queue, depending on the number of
available resources. If we also consider that assets may have varying levels of business
importance, the number of potential outcomes quickly becomes unmanageable. This
situation captures the problem we are interested in: given vulnerability, exploit and
outage events, an IT network, stated business needs for IT services and limited
resources, how can we decide which is the next action to take amongst options of
fixing, patching, isolating risk-exposed assets, or simply waiting?
2. Reinforcement Learning for Autonomic Computer Network Defence
In a decision tree such as the one previously introduced, assessing the cumulative risk
for every possible branch, in a greedy way, is not a practical strategy. We need to be
able to sample this state-action space and steer exploration towards the most promising
action selection strategy. This strategy is also known as a policy, and the optimization
goal we seek is to minimize business risk over a given time horizon. The search for
such an optimal policy is referred to as the Generalized Policy Iteration (GPI) problem,
and different approaches have been used to solve it, including Reinforcement Learning
[3]. Reinforcement Learning has been successfully used in complex control loop
systems, namely in automated packet routing problems [4] and automated server
resources allocation problems [1]. One of Reinforcement Learning’s benefits is that it
can be used online (adaptation to situations as they occur), offline (planning for
anticipated situations), on-policy and off-policy. The two last terms refer to whether the
learned policy is used to influence exploration (on-policy), or whether the policy
iteration is controlled through another mechanism independently, such as random
selection, human control, or heuristic rules (off-policy). Finally, Reinforcement
Learning can make use of state generalization methods such as function approximators,
to scale in continuous state space applications [5].
In Reinforcement Learning, an agent is rewarded after reaching a goal, and this

reward is discounted to each preceding action through exploration and policy iterations
(also known as epochs). Since our objective is to find a policy leading to the minimum
risk exposure, we considered both immediate risk reduction, ΔR(t), and the integral of
R(t) over the full simulation period, as potential reward functions. We then
experimented with various Reinforcement Learning algorithms such as Q-learning, and
parameters such as eligibility traces, learning rates, discount factors and random
exploration thresholds, to attempt to converge to a desirable CND policy.
3. Autonomic CND Experimentation Framework Architecture
Our experimentation framework is presented in Figure 2. It is broken down into seven

main modules, shown in dark blue, and which we describe in this Section.
Figure 2. Autonomic CND Experimentation Framework Architecture

Action Selection: For every CND environment state change, the Action Selection
module searches the policy for the best next action, action’, to implement given the
current state. This is performed either through a greedy search or using a softmax
Boltzman probability driven choice. The resulting action’ (fix, patch, isolate an asset,
or wait) is than passed to a resource and scheduled using Discrete Event Scheduling.
Discrete Event Scheduling: The scheduling module generates event duration and
puts the action event in a queue, ordered by time. In some scenarios, it also receives
exogenous events from the CND Environment Stochastic Model. The scheduling
module then advances the simulation clock to the next event in the queue and
communicates the associated State variables changes to the CND Graph model. We
implemented this module using the DEDS Java library called ABCMod from
University of Ottawa [6], which also supports random seed management and sample
dataset collection.
CND Environment Stochastic Model: This module generates arrival times for
outages, vulnerabilities and exploits according to predetermined probability
distributions. Leaving the details of these distributions to other forums, we used
Poisson distributions, implemented using the CERN Colt Java library 2 , with the
following means parameter λ values: 0.0134 vulnerabilities per hour, 0.0093 exploits
per hour per vulnerable/exposed host, and 0.00036 outages per hour per host (which
includes maintenance related outages). These values were derived from a simple
statistical and empirical analysis of incident reports3 and public data sources4. Details
can be found in [7] and [8].
CND Graph model: The CND Graph model keeps track of the status of each
assets, their interdependencies, and their support to the business processes (needs).
This module also enforces the rules for asset status changes (OK, vulnerable, outage or
exploited), considering safeguards, actions and exogenous events. This model was
implemented using the JGraphT5 Java library.
Risk Assessment: The Risk Assessment module queries the CND graph Model for
the list of affected assets, computes the instantaneous risk R(t) and its integral over the
simulation run, then passes these scalars to the RL algorithm. The risk is updated
periodically and considers cumulative effects of potential and actual damages. The
dynamic risk assessment algorithm is shown in Eq. (1). Its details are kept to other
forums [].
n m
R (t ) = ∑ ∑ d i (t ) * p j (t )
i =1 j =1
(1)
Where:
• n is the number of affected assets;
• m is the number of events;
• di(t) is the damage function incurred by the business at time t for asset i;
• pj(t) is the likelihood of occurrence of an exploit for a vulnerability event
j at time t; 1 otherwise.
RL algorithm: Before each action implementation is completed, the RL algorithm

queries the CND Graph Model for the current state. After the action implementation,
the RL module queries the Graph Model for the new state’ and updates the policy with
these quantities and the associated reward ΔR(t), or the integral of R(t), depending on
the training strategy. This module is implemented using the QConnectionist [9] Java
package6, and University of New South Whales Reinforcement Learning Java package
by Time Eden, Anthony Knittel and Raphael van Uffelen7 .
Policy: The policy updates its state-action map with rewards and states received
from the RL algorithm. It also provides the action selection module with the preferred
action for a given state. This was implemented using the neural network provided in the
QConnectionist framework [9] for the state generalization policy, and standard Java
vectors objects for the table policies.
2
The Colt Java library is available at http://acs.lbl.gov/~hoschek/colt/
3
Trouble Tickets recorded at the Canadian Forces Network Operations Centre
4
Primarily from the National Institute of Standards and Technology (NIST)
5
JGraphT is an open source project available at http://jgrapht.sourceforge.net/
6
The QConnectionist source code is available at http://www.elsy.gdan.pl/
7
Source available at: http://www.cse.unsw.edu.au/~cs9417ml/RL1/applet.html
4. Experiment
We implemented a simple CND environment model, as shown in Figure 3. It includes

eleven nodes regrouped under four interconnected sites in order to create multiple
service paths. One site hosts a DNS server, one provides access to the internet gateway
(Router-4), and two other sites host each an email server and a client. The business
needs include email communications between Email-2 and Email-5, as well as
browsing the internet from Browser-3. Functional dependencies exist between the
email clients and the email servers, as well as between the browser and the DNS. These
business needs and functional dependencies form logical dependencies between assets,
shown by orange arcs. Using this model, we ran five simulation experiments: fixing
four concurrent outages, fixing eleven concurrent outages, patching eleven concurrent
vulnerabilities, patching a vulnerability or fixing exploits, and patching-fixing-
isolating-waiting in a continuous event arrival simulation. We used five different
policies for each experiment: reinforcement learning with a table and a neural network,
as well as heuristic policies including fixing or patching assets randomly, fix or patch
the asset with the highest value first, and doing nothing.
Figure 3. Simple CND environment with eleven inter-dependant assets
For each simulation, random number generator seeds for DEDS event timings
were managed to assure independence of results, avoid over fitting local timing
conditions and allow policy comparisons.
5. Results and Discussion
Prior to running simulations, we computed the contribution of each asset to the stated
business needs. This was accomplished using an asset valuation algorithm based on the
ratio of business-enabling network paths supported by each asset described in [8]. In
our simple CND environment, we found sixty-six paths through greedy, depth-first,
search. The resulting asset values, presented in Table 1, were later used as damage
metrics by the dynamic risk assessment module.
Asset name Asset Value Stated business needs

Server-1 0.7
Email-2 0.7 0.3
Browser-3 0.3 0.3
Router-4 (WWW) 0.3
Email-5 0.7 0.4
Server-6 0.7
DNS-7 0.3
Router-8 0.72
Router-9 1.0
Router-10 0.72
Router-11 0.88
Table 1. CND environment asset value results
We then conducted simulation runs for each scenarios, which we repeated in the
form of epochs to train our RL policies. In the first case, the learning task was to fix
four concurrent asset outages in an optimal sequence to minimize risk (risk equal
damages in this case, since p=1 for outages). Both RL policies converged to the same
risk performance, which was lower than the three other heuristic policies, as shown in
Table 2.
Policy Integral of R(t)

Let risk grow 41.4
Fix random 31.73438
Fix highest 17.5375
Q-Connectionist 16.3825
Q-Learning RL Table 16.3825
Table 2. Risk integral results of different policies for fixing four concurrent outages.
A sample of the exploration of the solution space by the RL agent can be seen in
Figure 4. The agent initially explored, rather randomly, various actions leading sparsely
distributed risk results, than converged to action sequences optimizing its reward
(minimizing risk in this case). We notice a trend around 32, which is the result for the
random action sequence used for exploration. Any results larger than 32 were caused
by choosing “wait” actions. These choices were punished through the risk reward and
became less frequent as training progressed. We can also observe a secondary periodic
value after convergence at around 17. This value corresponds to fix or patch the asset
with the highest value first, which is an expected equilibrium since asset values
contribute largely to the risk rewards for this scenario. The upper bound of this graph is
approximately 41 and corresponds to waiting for the entire simulation period.
Figure 4. Exploration and convergence pattern for a neural network Reinforcement Learning agent
The other scenarios trialed had significantly larger solution spaces and statistical
analysis was required to analyze the results. For this purpose we used Chi-Square
metric with the Student distribution and we adjusted the number of runs to maintain our
results within approximately 10%, as a quality measure. In all scenarios, the policy
learned by the Reinforcement Learning agents achieved lower risk in a statistically
significant way when compared to the random policy. These results are shown in Table
3.
QConnectionist
Q-Learning
Random Improvement Neural Network Improvement
Scenario Table policy
(avg risk) (avg risk) policy (avg risk)
(avg risk)
(avg risk)
1. Fix 4 outages 31.73 16.38 -15.35 16.38 -15.35
or wait.
2. Fix 11 outages 17.38 16.45 -0.93 13.42 -3.96
3. Patch 11 1.87 1.60 -0.27 1.77 -0.10
vulnerabilities
4. Patch 1 1.88 1.61 -0.27 1.81 -0.07
vulnerability or
fix exploit
5. Patch, fix, 18933 18402 -531 18327 -606
isolate, or wait
with continuous
event arrivals
Table 3. Reinforcement Learning policies results compared against the random policy.
In Figure 5, we show a sample simulation run for scenario 2, where 11 concurrent

outages had to be fixed in an optimal sequence. The graph presents various decision
points and the risk function R(t) for the five trialed policies. Because the random seed
is the same for all actions timing in this case, we observed that some policies were
more effective at finding root-cause outages early, hence lowering their overall risk
score.
Policy Comparison over 1 Simulation Run

1.2
Qconnectionist NN policy
0.8 Fix Random
Fix highest asset first
R(t)
0.6 Let risk grow

Action completed
0.4 Q-learning with table policy
0.2
0
0 5 10 15 20 25 30 35 40 45
Time (Hours)
Figure 5. Policy comparison for fixing eleven outages in a single simulation run.
Over a 10-year continuous simulation, we observed that policies may be locally

optimal for risk, but globally very poor. Figure 6 shows the last 15 days of such a
simulation run. The turquoise fill represents the random policy area under R(t).
Although, the random policy had the highest risk over the full simulation period, it
achieved local optimums in regions marked by the red dashed ovals in the graph.
Figure 6. Comparison of policies over fifteen days of a 10-year continuous simulation.
In our experiments, the Reinforcement Learning agents seemed capable of learning

generally good policies, but they could not account for all possible situations caused by
the various event arrivals and action implementation delays. Even for our simple CND
environment, in order to achieve globally optimal policies, more information may have
been required (using a different CND state representation, as an example) as well as
different learning strategies (more epochs and different parameters).
We were finally interested in evaluating the effect of resources on risk and policies.
We conducted a number of 1-year simulation runs, with increasing numbers of NOC
staff resources, for two different policies: fixing or patching assets randomly and fixing
or patching the highest asset value first. The averaged results are shown in Figure 7.
Convergence of different policies as resources increase

( 1 year continuous simulation)
10000
Random policy
Highest Value First policy
Log scale of the integral of R(t)
1000
100
0 1 2 3 5 10
Number of NOC staff
Figure 7. The effect of resource availability on overall risk.
We can observe that as the response capacity (number of NOC staff) is increased,
not only is the integral of R(t) decreasing, but so is the difference between both
policies 8 . Indeed, if there are no resource constraints, there is no need to prioritize
responses since there is no event queue, and decision-making becomes essentially
trivial. This observation clearly supports the value of automation in CND.
6. Conclusion and Future Work
In this research effort, we applied Reinforcement Learning to the problem of finding an

optimal policy for Autonomic Computer Network Defence. We argued the need for a
controller able to dynamically iterate through various policies and retain the best
performing one. We have shown that risk maybe a good metric to steer such a
controller, as long as it accounts for actual and potential events. We presented our
experimentation framework and validated our concept using a simple CND
environment and five scenarios. Our results show that Autonomic CND using risk
states and Reinforcement Learning is possible, but that policies obtained, although
generally good, did not represent global optimums.
As future work, we propose investigating further dynamic risk assessment
algorithms and methodologies. We also suggest investigating different Computer
Network Defence state representations, for use in conjunction with Reinforcement
Learning agents herein tested, to see if better policies can be obtained. Namely, we
propose investigating text mining techniques, including feature extraction, and consider
modeling the CND environment as a “bag-of-words” to leverage these techniques.
Finally, we suggest looking into scalability issues, namely investigating distributed
8
Note that the upper bound of Figure 7 represents having no response capacity, which makes all policies
equivalent by default.
policies and the use of Collaborative Reinforcement Learning to achieve superior risk
results and stability.
References
[1] Tesauro, Reinforcement Learning in Autonomic Computing, IBM T.J. Watson Research Center, IEEE
2007.
[2] Benjamin, Pal, Webber, Atighetchi, Ruber, Automating Cyber-Defense Management, ACM Workshop on
Recent Advances in Intrusion Tolerant Systems, 2008.
[3] Sutton, Barto, Reinforcement Learning:An Introduction, MIT press, 1998.
[4] Boyan, Littman, Packet Routing in Dynamically Changing Networks: an RL approach, Advances in
Neural Information Processing Systems, Morgan Kaufmann, San Francisco CA (1993), volume 6, 671-
678.
[5] Sutton, McAllester, Singh, Mansour, Policy Gradient Methods for Reinforcement Learning with
Function Approximation, Advances in Neural Information Processing Systems 12, 2000.
[6] Birta, Arbez, Foundation on Modeling and Simulation, University of Ottawa, 2006.
[7] Alhazmi, Malaiya, Quantitative Vulnerability Assessment of Systems Software, Reliability and
Maintainability Symposium, 2005.
[8] Beaudoin, Japkowizc, Matwin, Autonomic Computer Network Defence Using Risk States and
Reinforcement Learning, Thesis manuscript to be submitted, University of Ottawa, 2009.
[9] Kuzmin, Connectionist Q-Learning in Robot Control Task, Riga Technical University, 2002.
[10] Cao, From Perturbation Analysis to Markov Decision Processes and Reinforcement Learning, DEDS:
Theory and Application, 2003.
[11] Chairman of the Joint Chiefs of Staff Instruction, Information Assurance and Computer Network
Defense, US DoD, 2004.
[12] Dobson, Denazis, Fenandez, Gaiti, Gelenbe, Massacci, Nixon, Saffre, Schmidt, Zabonelli, A survey of
Autonomic Communications, ACM Autonomous and Adaptive Systems, Vol. 1, No. 2, 2006.
[13] Kotenko, Multi-agent Modelling and Simulation of Cyber-Attacks and Cyber-Defense for Homeland
Security, IEEE International Workshop of Intelligent Data Acquisition and Advanced Computing
Systems, 2007.
[14] Lefebvre, Grégoire, Froh, Beaudoin, Computer Network Defence Situation Awareness Information
Requirements, MILCOM 2006.
[15] Moitra, Konda, The Survivability of Network Systems: An Empirical Analysis, CMU SEI, 2000.
[16] Ryan, iWar: A new threat, its convenience – and our increasing vulnerability, NATO review, 2007.
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-249
Enhancing Graph-based Automated DoS

Attack Response
Gabriel KLEIN a, Marko JAHNKE a, Jens TÖLLE a,1
Peter MARTINI b
a
Research Institute for Communication, Information
Processing and Ergonomics (FGAN-FKIE), Germany
b
Insitute of Computer Science IV, University of Bonn, Germany
Abstract. Timely and appropriate reactions to detected denial-of-service attacks

against computer networks are crucial in both civilian and military settings.
GrADAR is an intuitive graph-based approach for assessing the effects of DoS at-
tacks against computer networks so that response measures can be automatically
selected without human intervention. However, GrADAR has limitations insofar
as implicit effects of countermeasures are only taken into account by propagation
towards user nodes. Possible effects in the other direction are only considered if
they are explicitly specified. For this, they need to be exactly known in advance
which is often infeasible. This contribution presents an extension to GrADAR, in
which we consider resource workload and processing capabilities and their effects
on resource availability. We incorporate workload measurements into the GrADAR
model which are done by passive analysis of network traffic. We further augment
the active availability probes with passive measurements. This ensures more ac-
curate availability values because additional measurement traffic that might falsify
results only needs to be injected when resources are currently not accessed.
Keywords. denial-of-service attacks, automated response, response evaluation,
passive availability measurement
Introduction
In recent years, the number of attacks against computer networks has steadily increased.
They cannot only be observed in civilian scenarios (e. g. e-commerce or online banking)
but also in military settings. Among these attacks, denial-of-service attacks are the most
prevalent, often resulting in inaccessibility of services and/or entire networks. This can
result in enormous financial losses, in case of commercial applications, and negatively
impact battle readiness where military networks are concerned.
In typical wired networks, such attacks can be detected with a high degree of accu-
racy by incorporating intrusion detection systems into the networks’ perimeter defence.
Once detected, security personnel can then suitably react to these attacks. However, be-
cause of the increased speed and reliability, it is desirable to react to detected attacks in
an automatic fashion. For the automatic selection of response measures, it is necessary
1 Corresponding Author: Gabriel Klein, FGAN-FKIE, Neuenahrer Str. 20, 53343 Wachtberg, Germany;
E-mail: g.klein@fgan.de
250 G. Klein et al. / Enhancing Graph-Based Automated DoS Attack Response
to quickly and accurately estimate the effects of the respective countermeasure on the
network resources.
In previous work, we have proposed GrADAR [1,2], an intuitive approach to create
and maintain a model of a computer network and the availability of its resources from the
observations of deployed monitoring systems. The graph-based model is able to express
both the effects of DoS attacks and the results of available response measures prior to
their application in the real-world network. Thus, the approach provides a methodology
for automatically selecting response measures to detected attacks. The most appropriate
response is chosen based on metrics which are well-known from the pragmatic view of
network security officers.
This contribution proposes an extension to our previous GrADAR approach that
seeks to incorporate the effects of network and resource workload into the availability
estimation. This will permit a more detailed modelling of the current network state. Fur-
ther, it will allow the specification of the effects of more complex DoS countermeasures.
To further improve the availability measurements and reduce the workload placed on the
network and resources, passive measuring is employed instead of active probing.
The rest of this paper is structured as follows: section 1 introduces related work done
in the area of passive measurements. Subsequently, section 2 gives an overview of the
GrADAR approach as well as some of its limitations. Section 3 presents the proposed
GrADAR extensions, describing workload and the measurement framework in more de-
tail. Section 4 portrays first results and, following that, section 5 presents a summary and
an outlook on future work.
1. Related Work
The area of passive network analysis has been a research area of interest for a number
of years. In the context of this contribution, approaches concerned with the inference of
server workload and the identification of network flows are of particular interest.
In [3], Barford and Crovella describe an architecture for actively and passively mea-
suring the effects of Web server and network workload on the quality of client connec-
tions. They show that increasing network load causes a deterioration of connection qual-
ity. However, they observe a positive effect of server load on traffic characteristics and
attribute this to a reduced burstiness of traffic.
Eriksson et al. [4] reiterate a methodology for determining the network structure by
passive measuring of hop counts. They address the issues of missing hop count data and
how to extract topological information from large hop count databases.
Passive measurement is also a popular method for determining the characteristics
of network connections such as bandwidth, latencies or packet loss statistics. Seshan et
al. [5] propose SPAND, a system in which sensors distributed throughout the network
perform passive measurements and report to central performance servers where the data
is collated. In [6], Lowekamp presents a report on the Wren project which deals with
the development of network performance monitoring solutions. Here active and passive
measurements of traffic statistics are combined to reduce the amount of artificial traffic
wherever possible.
To passively determine availability and workload, observed traffic needs to be as-
sociated with the resources between which it flows. Distinct flows can, for example, be
G. Klein et al. / Enhancing Graph-Based Automated DoS Attack Response 251
identified using NetFlow [7], but correlations between within the same or different flows
also need to be ascertained subsequently. This is often done by simple payload inspec-
tion, although this potentially requires large amounts of memory. However, recent ap-
proaches for the identification of upper layer protocols include machine learning [8,9]
and multi-scale gamma models [10].
2. GrADAR Overview
In the GrADAR approach, a simplified model of the real-world network is created in or-
der to predict the effects of available response measures against denial-of-service attacks.
Figure 1 shows an overview of the approach.
Figure 1. Schematic overview of the GrADAR approach.
2.1. Nomenclature
The core concept of GrADAR is based on the availability of resources. Resources (as
suggested in [11]) can be either services provided by hardware or software components
(denoted as S), or users (denoted as U). Therefore, the set of resources is R = S ∪ U.
Each resource r has an associated value A(r) ∈ [0, 1], signifying the extent to which
it is available to other resources. In [12] and [13], the concept of resource-typical transac-
tions was proposed. We adopt this concept and define a resource’s availability as the time
needed for a transaction with the resource. Since a resource typically requires interaction
with other resources to function correctly, we assume that its availability is the result
of two independent factors, an internal state (the intrinsic availability) and the values of
other associated resources (the propagated availability). Thus, a resource’s availability

is defined as
A(r) = AI (r) · AP (r) (1)
for each r ∈ R.
A resource r may be dependent on other resources s1 , . . . , sn (denoted
as r s1 , . . . , r sn ). In this case, the degree to which r depends on each of these may
vary [14,15] and can be specified by weighting the respective dependency. This can be
formalised as
AP (r) = Dr (wr,s1 (A (s1 )) , wr,s2 (A (s2 )) , . . . , wr,sn (A (sn ))) , (2)
where Dr : [0, 1]n → [0, 1] is a dependency function and wr,si : [0, 1] → [0, 1] are
corresponding dependency weighting functions. In optimal conditions,
Dr (wr,s1 (1) , . . . , wr,sn (1)) = 1.
A more detailed discussion of this can be found in [2].
2.2. Dependency Graph
To represent the availability dependency relationships between the set of resources R, the
resources in the real-world network are modelled as a directed acyclic graph Ĝ = (V̂ , Ê)
with V̂ ⊆ R and Ê ⊆ ((S ∪ U) × S). Its vertices correspond to the resources and the
edges correspond to the dependency between the respective resources. These resource
dependencies need to be determined beforehand, either analytically or experimentally. Ĝ
contains an edge (r, s) iff rs. These edges are labelled with the corresponding weighting
function wr,s . This graph is called the dependency graph and reflects the ideal state of
the network. Let Ĝ be the set of all possible dependency graphs.
2.3. Accessibility Graph and Overall Availability
A DoS attack typically affects the availability of resources. Thus, there is the possibility
that some resources might no longer be accessible to others. Therefore, a second graph
is required, the so-called accessibility graph. Mutual accessibility of a set of resources R
is expressed by a graph G = (V, E) with V ⊆ R and E = ((S ∪ U) × S), in which an
edge (r, s) exists when a resource s is directly accessible from r. The vertices r ∈ R of
the accessibility graph are labelled with the corresponding resource’s availability A(r).
The availability of user nodes is interpreted as the user-perceived availability of the
network. Since the network supports one or more users or groups of users in conducting
a common mission, we define the overall availability of the network as the weighted
average of all user nodes’ availability values:

A(G) := m(u) · A(u),
u∈U
where m(u) is the relative importance of useru to the common mission which needs to
be determined beforehand or adaptively, and u∈U m(u) = 1. Let G be the set accessi-
bility graphs.
Usually, monitoring systems deployed in the network will only be able to observe
availability values for some of the network’s resources. This is especially true for users,
for which an availability cannot be objectively measured. Thus, the availability of re-
sources for which values cannot be observed need to be estimated. For a resource r with
r s1 , . . . , r sn , this estimation is done by propagating the availability values of the
resources s1 , . . . , sn in the inverse direction of the corresponding dependency relation-
ship expressed in the dependency graph, i. e. along the edges (s1 , r), . . . , (sn , r), and
then calculating A(r) according to equations (1) and (2) with AI (r) = 1.0. This can be
efficiently done, for example, with a depth-first search algorithm, starting from the user
vertices and terminating at vertices with degout (r) = 0.
As opposed to the dependency graph, the accessibility graph shows the actual current
state of the network.
2.4. Response Selection
Once an attack has been detected, an appropriate reaction should be selected automat-
ically. With the current dependency graph Ĝ and the current availability graph G, we
define a countermeasure or response measure as a transformation
θ : Ĝ × G → Ĝ × G.
The dependency graph Ĝ = (Vˆ , Ê ) is obtained from Ĝ by adding or removing vertices
or edges, and the accessibility graph G = (V , E ) is derived from G by also adding or
removing vertices or edges but, additionally, vertex availability values can be changed.
Let the set of available countermeasures be denoted as Θ.
As response measures may be arbitrarily complex in nature, we assume that a single
response measure θ can be divided into Nθ ∈ N+ successive atomic response steps θ(i) :
θ = θ(1) ◦ . . . ◦ θ(Nθ ) .
Each of these real-world response steps θ(i) corresponds to one of the graph transforma-
tion primitives mentioned above (adding/removing vertices or edges, setting availabil-
ity). The effects of such a response step can be either directly associated with the ac-
tion (explicit impact) or a result of changes in the environment due to the response step
application (implicit impact).
For the automatic response measure determination, each available countermeasure
θ ∈ Θ is now applied in parallel to the current accessibility graph. For each change in the
accessibility graph, the propagation algorithm mentioned above needs to be executed.
The resulting graphs (so-called response graphs) are then compared with respect to dif-
ferent metrics, e. g. expected response success or expected response costs [2], and the
most appropriate response is then applied to the real-world network.
The resulting dependency graph of one such GrADAR cycle is used as the input
for the subsequent iteration. Thus, the process constitutes a so-called closed-loop control
system.
2.5. Current Limitations
During the validation of the GrADAR approach it became apparent that correctness and
robustness could only be achieved if the precise effects of real-world countermeasures
in the graph space could be accurately predicted. Because of the closed control loop
structure of the approach, subsequent iterations of the loop would operate on incorrect
input.
So far, the response measures for which effects were specified consisted only of
blocking access to specific resources (e. g. closing a firewall port) or migrating resources
to different locations (e. g. installing a new server to replace another). The effects of these
operations on the graph could be specified fairly easily in terms of changed availability
values or changing edges in the graph.
However, more complex behaviour of resources, such as the interactions between
workload placed on a resource and its processing capacities and their effect on that re-
source’s availability, are only expressible if fairly well known and specified in advance.
This is because the effects are taken into account by propagating them through the acces-
sibility graph according to the update algorithm. This imposes constraints in that effects
can only be propagated in the reverse direction of the dependency relationship. Effects on
resources in the other direction cannot currently be expressed. For example, blocking a
DoS attack against a Web server at a firewall port would have the explicit impact of a 0.0
availability at the firewall port. This would be propagated to the (dependent) user node.
However, a possible implicit impact could be an increased availability of the Web server.
This is currently expressible only if the changes in availability are known beforehand
and “hard-coded” into the graph transformation, something which is often impossible.
It is desirable to predict such effects on resource availability dependent on the current
network situation.
Furthermore, availability measurements are currently performed only through active
probing, i. e. by sending requests to the respective resources and measuring and normal-
ising the time required for an answer. This poses two problems. First, the measuring
process itself produces workload for the network and the target resource, and second, it
requires the measuring process to produce a traffic pattern which is representative for the
specific resource.
3. Beyond GrADAR: Improving Availability Estimation
Due to the limitations recounted in section 2.5, we propose to enhance the GrADAR
approach by incorporating the effects of workload and resource capacity into the graph-
based model and avoiding active probing in favour of passive availability measurements.
We believe that through the enhancements described in this section the correctness and
robustness of the approach can be significantly improved.
3.1. Solution Idea: Propagation of Workload
To more accurately represent the graph-based equivalents of DoS attack response mea-
sures, we propose to incorporate the workload placed on the various resources into the
GrADAR model. For this, the relationships between resource workload and its availabil-
ity need to be determined, especially in the area of a resource’s processing capacity. Ad-
ditionally, dependency relations between the workload of different resources should be
investigated. If such relationships exist, this could allow the inference of workload values
for resources of which workload cannot be directly measured. Similar to availability es-
timation, this could be done by propagation according to the workload relationships; see
figure 2 for an example. Here, two users (or user groups), User_1 and User_2, com-
municate with an HTTP server, HTTP_S1, via two separate firewall ports, HTTP_F1
and HTTP_F2. In figure 2(a), the workload generated during a DoS attack by User_1
is propagated (in the direction of the dependency relationship) to the firewall port and
the HTTP server. Because of this workload, the HTTP server has an availability of 0.0
which, in turn, is propagated (against the dependency relationships) to the resources de-
pendent on it. An example response to such an attack, namely blocking the attacker at
the firewall, is depicted in figure 2(b). As a result, the workload generated by User_1 is
no longer propagated to HTTP_S1 which causes an increase in the server’s availability.
This, in turn, is propagated to User_2.
(a) Workload-based DoS attack. (b) Example response to DoS attack.
Figure 2. Example of workload propagation in the GrADAR model.
3.2. Workload Definition
Before the workload L(r) of a resource r ∈ R can be effectively measured, it needs

to be defined in a suitable fashion. The dictionary defines workload as “the amount of
work assigned to, or done by, a worker or unit of workers in a given time period” (The
American Heritage Dictionary, 2nd Edition). Similar to the definition of availability as
the normalised duration of a resource-typical transaction (see e. g. [12,13]), a resource’s
workload can thus be defined as the number of typical transactions a resource needs
to process per unit of time. Since a networked application scenario contains multiple
types of resources, different types of “work” need to be considered as each resource has
transactions that are typical for it; for example, the number of concurrent transactions a
Web server needs to process. Table 1 contains a listing of possible resources along with
the availability and workload definition for each of them.
Table 1. Workload definitions for selected resources.

Resource Availability definition Workload definition
IP stack ICMP ping response time IP packets/time
CMS Delay for receiving backend content Current active transactions
IRC server Delay for connection, joining channel and Current active transactions
sending a message
DB server Delay of query from Web server backend Current no. of transactions
DNS server Delay for result of lookup query Requests/time
MAC layer Interface up/down Frames/time
CPU Average CPU load
Execution delay for application requiring
Memory Average memory consumption
CPU/memory/HDD
HDD Average consumed HDD capacity
To adequately compare the workload of different types of resources, workload val-

ues need to be normalised:
L(r)
L̃(r) = , (3)
Lmax (r)
where Lmax (r) is the maximum workload which a resource can adequately process
within a certain time frame. This is closely related to the definition of availability (c. f.
[1,2]) where request-response delays are normalised with respect to a maximum accept-
able time from a user’s perspective.
3.3. Measurement Framework
The measurement of availability and workload is performed according to the framework

outlined in the SDL [16] diagram depicted in figure 3.
Passive sensors at appropriate locations in the network (e. g. at central switches or a
firewall) constantly observe passing traffic. Different conversations between consumers
and providers of a service, and the transactions they comprise are identified by analysing
packet headers and correlating certain fields, e. g. sequence numbers, IP addresses or
port numbers. For each recognised transaction, various properties of the traffic such as
average packet loss, transaction duration, average round-trip time, jitter, etc. can be used
to evaluate the availability of a resource; c. f. [12] and [13] for details of how traffic
properties can be used to make quantitative statements about the quality of a service.
Thus, a resource’s availability value can be updated after each completed transaction.
By logging the number of transactions for different services and/or protocols, these
sensors can also establish a current workload for the observed resources. As already
mentioned (c. f. table 1), the number of typical transactions within a specified period of
time (or possibly the ratio of initiated vs. completed transactions) can serve as a workload
metric.
The goal of our work is the identification of current threats to a network and the
ability to react in near-real-time. Bearing this in mind, it seems advisable to consider the
development of resource availability and workload over a customisable period of time
rather than only the currently measured values. Using this as a base for decisions may
reduce the likelihood of overreactions or false positives.
Figure 3. SDL diagram of the measurement framework.
The aforementioned description of passive measurements assumes that representa-

tive traffic for all relevant resources can always be observed. This may not be the case
in real-world configurations, e. g. if clients are under EMCON in military settings. Thus,
to always retain an overview of the current network status, active measurements need
to be performed if no traffic was observed for a certain amount of time. In the case of
availability, this is done via an active probe which consists of a (representative) request
to the respective resource. The resulting availability is the normalised duration of this
request. The resources’ workload needs to be queried directly via appropriate interfaces
at the resources themselves (e. g. via SNMP [17]), although this introduces a certain de-
gree of dependency on specific applications and protocols. Note that, when performing
active measurements, the load generated by these artificial requests needs to be taken
into account and compensated when determining resource workload.
We are aware that both active and passive measurements have disadvantages. In the
case of active probing this is the injection of additional traffic into the network resulting
in increased workload for the targeted resources. On the other hand, when observing
traffic passively, the volume of traffic to be dealt with is potentially prohibitively large.
We try to avoid this by restricting ourselves to the analysis of only packet headers instead
of entire packets.
4. Preliminary Results
We have performed first simple workload and availability measurements in a network

topology depicted in figure 4. A client accesses a content management system (CMS)
which, in turn, retrieves its data from a database server on a different host (scenario 1). In
a second measurement scenario (scenario 2), both the CMS and the database are located
on the same host. In both cases, the servers are separated from the client by a firewall host.
The servers hosting the Web and database servers are older-model single core machines
and are thus not able to process requests entirely in parallel.
The values shown in figures 5 and 6 were obtained by generating an increasing
number of concurrent requests per time frame. For each number of concurrent requests,
the measurement process was repeated 300 times. The mean request-response durations
were used as the basis for the availability calculations.
DB
CMS
backend
Firewall
Client
Figure 4. Server setup during measurements.
Figure 5 shows the measurement results for scenario 1. The normalised CMS work-
load and availability are plotted against an increasing number of concurrent requests di-
rected at the Web server. At first, the CMS availability degrades linearly with the in-
crease in workload. Beyond around 15 concurrent requests, an overload situation is en-
tered, in which the availability remains at zero (shown as triangle-shaped data points in
the diagram). At this point, the server cannot process requests within an acceptable time
frame.
CMS availability
CMS workload
1.5
1.5
CMS overload (workload above 1.0)

CMS workload/availability
1.0
1.0
0.5
0.5
0.0
0.0
0 5 10 15 20 25
Number of concurrent requests
Figure 5. CMS workload and availability plotted against increasing number of concurrent requests; CMS and
DB server on separate hosts.
In the second scenario, where the DB server is on the same host as the CMS server,
the CMS availability degrades slightly more quickly (depicted in figure 6). This is most
probably because both server processes share the same system’s resources. Load pro-
cessed by the CMS is partially transferred to the database, which, in turn, reduces the
CMS availability
CMS workload
1.5
1.5
CMS overload (workload above 1.0)
CMS workload/availability
1.0
1.0
0.5
0.5
0.0
0.0
0 5 10 15 20 25
Number of concurrent requests
Figure 6. CMS workload and availability plotted against increasing number of concurrent requests; CMS and
DB server on the same host.
available resources for the CMS. Also, the availability curve is less smooth. The small
confidence intervals suggest that the reasons for the outliers are systematic. They could,
for example, be caused by changes in scheduling policies. This is also true for the outlier
observed in scenario 1.
1.0
1.0
0.8
0.8
CMS availability
CMS availability
0.6
0.6
0.4
0.4
0.2
0.2
0.0
0.0
0.0 0.5 1.0 1.5 0.0 0.5 1.0 1.5
CMS workload CMS workload

cor. coeff.: −0.934 cor. coeff.: −0.934
(a) Separate CMS and DB servers (b) Common CMS and DB server
Figure 7. Relationship between resource workload and availability.
Figure 7 shows the relationship between workload and availability for both scenar-
ios. In both cases, the correlation coefficient is very close to −1. This suggests the ex-
istence of a functional dependency between resource workload and availability. In this
simple first example, we observe a linear dependency between the two values. However,
more complex dependencies may exist, e. g. in the case of a multi-core system which is
able to process multiple requests in parallel. Here, up to a certain workload, the avail-
ability should not be markedly impaired at all. Also, when considering other types of
resources (e. g. the operating system kernel), binary relationships are possible, where a
resource remains fully available up to a certain workload, after which it immediately

drops to zero.
5. Summary and Further Work
This contribution has discussed an extension to GrADAR, an approach for automatically

assessing the effects of denial-of-service countermeasures. Including the effects of re-
source workload into the GrADAR model permits the specification of complex counter-
measure effects.
Workload and availability measurements were performed in a simple DMZ-like
setup which included machines representing a client, a firewall host and two servers.
They were conducted using a passive monitoring solution capable of calculating resource
workload and availability from observed network traffic. The results of these measure-
ments indicate a possible functional dependency between resource workload and avail-
ability which justifies the incorporation of workload measurements into the GrADAR
approach.
The work done regarding the GrADAR extension is of a preliminary nature. There
are numerous aspects which need to be considered in future work. So far, the generated
traffic used for measuring workload and availability consisted only of a steadily increas-
ing number of concurrent clients requesting the index page of an e-commerce Web site.
Representative traffic for such a scenario needs to be generated for a more detailed eval-
uation; e. g. according to a formal customer state model as depicted in figure 8 with dif-
ferent states for each type of viewed page and appropriate state transition probabilities.
Also, possible dependencies between the workload of different resources needs to be
investigated, e. g. workload placed on the CMS and its backend database.
P1,4
P5,6
Main View View View Check−
page Search
article recomm. cart out
1 2 3 4 5 6
P5,2
Figure 8. Possible state model underlying browsing by Web shop customers.
Where the passive analysis of traffic is concerned, problems may arise when only
parts of conversations between resources can be observed, e. g. due to node movement
in mobile ad hoc networks. In this case it might become necessary to harmonise the
observations of multiple sensors distributed throughout the network.
Acknowledgements
The authors wish to thank Christian Thul for his excellent development work and his
valuable feedback.
References
[1] M. Jahnke, C. Thul, and P. Martini. Graph based metrics for intrusion response measures in computer
networks. In Proc. of the 3rd LCN Workshop on Network Security. Held in conjunction with the 32nd
IEEE Conference on Local Computer Networks, Dublin, Ireland, October 2007.
[2] M. Jahnke, C. Thul, and P. Martini. Comparison and improvement of metrics for selecting intrusion
response measures against DoS attacks. In A. Alkassar, editor, Proc. of the Sicherheit2008 Conference,
Saarbrücken, Germany, April 2008.
[3] P. Barford and M. Crovella. Measuring Web performance in the wide area. SIGMETRICS Performance
Evaluation Review, 27(2):37–48, September 1999.
[4] B. Eriksson, P. Barford, R. Nowak, and M. Crovella. Learning network structure from passive mea-
surements. In IMC ’07: Proceedings of the 7th ACM SIGCOMM Conference on Internet Measurement,
pages 209–214, New York, NY, USA, 2007. ACM.
[5] S. Seshan, M. Stemm, and R. H. Katz. SPAND: Shared passive network performance discovery. Tech-
nical Report UCB/CSD-97-967, EECS Department, University of California, Berkeley, August 1997.
[6] B. B. Lowekamp. Combining active and passive network measurements to build scalable monitoring
systems on the grid. SIGMETRICS Performance Evaluation Review, 30(4):19–26, 2003.
[7] B. Claise. Cisco Systems NetFlow Services Export Version 9. RFC 3954 (Informational), October 2004.
[8] N. Williams, S. Zander, and G. Armitage. Evaluating machine learning algorithms for automated net-
work application identification. Technical Report 060410B, Centre for Advanced Internet Architectures
(CAIA), March 2006.
[9] P. Barlet-Ros, V. Carela-Español, E. Codina, and J. Solé-Pareta. Identification of network applications
based on machine learning techniques. In TNC 2008: Proc. of the Terena Networking Conference, 2008.
[10] Y. Himura, K. Fukuda, K. Cho, and H. Esaki. Characterization of host-based traffic with multi-scale
gamma model. In Proc. of the 2nd CAIDA/WIDE/CASFI Workshop, Seoul, South Korea, April 2009.
[11] T. Toth and C. Kruegel. Evaluating the impact of automated intrusion response mechanisms. In ACSAC
’02: Proceedings of the 18th Annual Computer Security Applications Conference, page 301, Washing-
ton, DC, USA, 2002. IEEE Computer Society.
[12] J. Mirkovic, P. Reiher, and A. Hussain. Measuring denial of service. In Proc. of the ACM Workshop on
Quality of Protection (QoP), pages 53–58. ACM Press, 2006.
[13] J. Mirkovic, A. Hussain, B. Wilson, S. Fahmy, P. Reiher, R. Thomas, W. Yao, and S. Schwab. Towards
user-centric metrics for denial-of-service measurement. In Proc. of the Workshop on Experimental Com-
puter Science, 2007.
[14] S. Bagchi, G. Kar, and J. Hellerstein. Dependency analysis in distributed systems using fault injection:
Application to problem determination in an e-commerce environment. In Proc. of the 12th Intl. Work-
shop on Distributed Systems: Operations & Management, 2001.
[15] P. Bahl, P. Barham, R. Black, R. Chandra, M. Goldszmidt, R. Isaacs, S. Kandula, L. Li, J. MacCormick,
D. Maltz, R. Mortier, M. Wawrzoniak, and M. Zhang. Discovering dependencies for network manage-
ment. In Proc. of the V HotNets Workshop, 2006.
[16] ITU. Specification and Description Language (SDL) (ITU-T Recommendation Z.100). International
Telecommunications Union, August 2002.
[17] J. D. Case, M. Fedor, M. L. Schoffstall, and J. Davin. Simple network management protocol (SNMP).
RFC 1157 (Historic), May 1990.
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-262
On nth Order Attacks

Daniel BILAR a,1
a
Department of Computer Science, University of New Orleans, USA
Abstract. An nth order attack seeks to degrade, disable or subvert an end system
indirectly by targeting one or more end mission-sustaining ancillary systems. We
discuss the vulnerability etiology enabling such attacks. We illustrate the notion of
these attacks with concrete historical, current and forward-looking examples; also
in the context of cyberwar against advanced computerized societies. We sketch the
challenges and requirements to detect and mitigate the effects of nth order attacks.
Keywords. nth order attack, Highly Optimized Tolerance, ancillary system,
assumption violation, economic warfare, critical infrastructure
1. Introduction
The goal of nth order cyber-warfare is to induce instabilities in mission-sustaining an-

cillary systems that ultimately degrade, disable or subvert an end system. Such systems
may be technical/algorithmic; however, societal, psychological, ideological, economic,
biological and natural systems may be targets, as well. Ancillary systems include pars
pro toto memory resource allocation, throughput control, hardware/software manufactur-
ing, visualization environments, social welfare systems, human networks, power gener-
ation/transmission/distribution, voting systems, data and goods supply lines, reputation
management, entropy externalization, business models and economic systems.
For example, a denial of service attack against a web server can be seen as a case
of a 2nd order attack against the resource allocation subsystem of the TCP transport
subsystem. Thompson’s trojaned compiler in “Reflection on Trusting Trust” may be seen
as a 3rd order attack against software manufacturing tools [1].
This paper defines and discusses this class of attacks and tries to explain their etiol-
ogy via reference to Highly Optimized Tolerance (HOT) processes. HOT processes in-
duce structured systems through optimization mechanisms that incorporate tradeoffs be-
tween objective functions and resource constraints in probabilistic environments. Perti-
nent to our discussion is the property that such optimization-generated systems are robust
towards common perturbations, but especially fragile towards rare events, such as unan-
ticipated changes in the environment. Inducing such ‘rare events’ in mission-sustaining
ancillary systems is thus the goal of nth order attacks.
The rest of this paper is organized as follows: Sec. 2 explains the main concepts
that motivate our subsequent discussion. Sec. 3 reviews related work. We give concrete
examples of nth order attack in Sec. 4. Sec. 5 discusses analytical aspects of nth order
1 Corresponding Author: Daniel Bilar, Department of Computer Science, University of New Orleans, 2000
Lakeshore Drive, New Orleans LA 70148, USA; Email: daniel@cs.uno.edu

D. Bilar / On nth Order Attacks 263
attacks. Sec. 6 briefly sketches theoretical and practical remediation approaches. Sec. 7
gives final thoughts on the urgency of addressing the theme of the paper.
2. Overview
The following section serves to flesh out the nomenclature and concepts used throughout
the paper. We shall start with the abstract notion of a ‘system’; the definition of which
varies across time and domains. For the purposes of this discussion, we adopt a recursive
variant of biologist von Bertalanffy’s seminal work on General Systems Theory [2]:
A system is a whole that functions by virtue of the interaction between constitutive com-
ponents. As such, it is defined by these relationships. Components may be other systems.
For our purposes, the attractiveness of the definition lies in its emphasis on openness and
the allowance for structural similarities across different domains with concomitant corre-
spondence of governing behavior. For an short, readable, largely non-technical overview
of competing system theories, the reader is referred to [3, ch. 2].
Figure 1. System view end system
Ancillary systems are responsible for control mechanisms, fault detection/resilience/re-

covery, energy/data flow, economic viability, human usability, data processing/structures,
graceful startup/shutdown, reputation management, governance, social order and more.
Such systems may be technical/scientific/algorithmic; however, societal, psychological,
ideological, economic, biological and natural systems are included, as well.
Ancillary systems span different scales and varying orders of complexity. They may be
embedded in or encompass the end system, and may in turn be composed of and influ-
enced by other ancillary systems. Figs. 2(a) and 2(b) list an embedding (say a business
model) and embedded ancillary system (in this example human operators) with reference
to an end system (denoted by the center star) from Fig. 1.
264 D. Bilar / On nth Order Attacks
(a) Embedding System: Business Model (b) Embedded System: Human Operator
Figure 2. Examples components of embedding system (a) embedded system(b) of an end system
A Network Intrusion Detection System (NIDS) may serve as an illustrative end sys-
tem example. Its ancillary control system negotiates the data and instruction interplay
between sensors, analysis, database and decision/response engine. The ancillary visual-
ization system displays the events and possible remediation options. The human operator
subsystem must interpret the happenings and subsequently make the decisions that are
not automated to the best of its reasoning ability. The entropy externalization subsys-
tem is (among other things) responsible for cleaning out accumulation of dynamic data
through sliding windows/logging and filtering out sensor noise. The end system itself is
embedded in a business model that governs aspects of its design, implementation and
activity: profit model, signature update cycles, customer support and more.
The ancillary systems of the NIDS end system have themselves subsystems: Human
operators (Fig. 2(b)) field a visual subsystem subject to parameters (no UV sight, cer-
tain percentage of color-blindness, angular resolution etc). Their control system may be
thought of as their reasoning strength and limitations (cognitive dissonance, herd instinct,
unconscious intelligence [4] etc), as well as their physiological mechanisms (hormone
secretions of the hypothalamus that regulate sleep, hunger, temperature etc). The human
subsystem of human operators may be coworkers, friends, the fellow polity, family. En-
tropy externalization systems manifest themselves in physical (as in human waste prod-
uct expulsion), as well as mental and psychological mechanisms (stress relief through
exercise, keeping a diary, art, talking on the phone etc).
The business model (Fig. 2(a)) is embedded itself in an economic environment, say
a free market economy, which influences its setup (tax codes, corporate structure, sales
channels, liquidity parameters such as interest rates which determine acceptable debt ra-
tios etc). The control subsystem may consists of corporate governance, union influence,
mission statement, and legislative regulations. Its visualization subsystem may include
accounting publication systems (standardized formats like IFRS with its own assump-
tions), dress codes, as well as marketing approaches (corporate image, advertisements
etc). Human operator system may be stockholders, consultants writing the business plan,
company workers, product consumers, company management, and competitors. Finally,
the entropy externalization ancillary system of the business model may include mech-
anisms to off-set losses to subsidiaries, third-tier rebranding of products for steep sales
discount, ‘poison pills’ to counter hostile takeovers, corporate fusion plans, and more.
2.1. nth order attacks
An nth order attack tries to indirectly degrade, disable or subvert an end system by tar-
geting one or more ancillary systems.
With this qualitative definition in hand (which we shall pick up in Sec. 5), let us revisit
the NIDS example in Fig. 1 with its control, human, entropy externalization, embedding
and visualization ancillary systems. How would one go about perpetuating an nth order
attack against an NIDS? One could take on the control system via a DoS attack against
the response/decision engine, or try to supply fake/poisoned data to the analysis engine.
Given biological, cognitive and psychological human parameters, enough false positives
at 3am will make human operators tone down the sensitivity of the analysis component.
One form of entropy electronic equipment produces is heat. The vast majority of Intel
and AMD CPUs, for instance, reach critical heat at about 55-85◦C[5, p. 5-13], which may
cause the BIOS to shutdown to prevent damage: Hence, one attack against this entropy
externalization system raises the ambient temperature of the building in which the NIDS
components are deployed (say by low-tech clogging the climate intake vents). PNNL’s
Starlight [6] offers a comprehensive NIDS visualization system, replete with 2-D and 3-
D multimedia visualizations supporting comparisons and emphasizing interrelationships.
As can be intuited by the Starlight Network Intrusion Detection Graph2 in Fig. 3, once
data flow reaches a critical mass (by virtue of screen resolution and human limitations)
visuals will degenerate into saturated pixel blobs, obviating their usefulness. The suscep-
tibility of security visualization methods to intentional noise remains a serious concern,
as described by [7].
Why do these attack work? Why does any attack, cyber- or otherwise, work? The
answer we propose is surprisingly simple: Attacks work because they violate assump-
tions. Any finite system by design must incorporate implicit and explicit assumptions
into its structure, functionality, and language. These systems are formulated with ‘ex-
pected’, ‘typical’ cases in mind and the assumptions reflect these expected use cases: A
man-in-the-middle attack violates the assumption that you are talking to the party you
expected; a race condition attack violates ordering assumptions; a buffer overflow attack
violates an explicit resource assumption; BGP routing and DNS case poisoning attacks
violate implicit trust assumption of non-malicious open architecture participants. Like-
wise, terroristic activities in open societies are easy to pull off because spaces are open,
population freedom of movement not controlled - hence they violate implicit societal
trust assumptions. Lastly, many democratic voting schemes assume ‘honest’ voters, and
hence can be undermined by strategic voting [8]. There are scores of examples, in every
domain.
We shall revisit the trust assumption in open societies in Sec. 7. Our next goal, how-
ever, is to gain some intuition about the etiology of the problem: We present a putative
generative mechanism which crucially depends on assumptions to highlight the conse-
quences of violating said assumptions.
2 In the interest of fairness, it should be noted that this image is originally in color, not gray shades.
Figure 3. Starlight NIDS Graph
2.2. Highly Optimized Tolerance
Highly Optimized Tolerance (HOT) is a generative mechanism that seeks to explain the
structure, statistics and resiliency of interconnected systems. Originally proposed to ac-
count for the ubiquity of so-called power laws in natural and engineered systems, it has
been fruitfully applied to the study of forest ecosystems, router network robustness, inter-
net traffic, power systems and immune systems. The strength of HOT models is four-fold:
First, by virtue of its emphasis on evolved and engineered complexity through feedback,
tradeoffs between objective functions and resource constraints in a probabilistic environ-
ment, it models the majority of real-life systems which are subjected to such pressures.
Secondly, its features include high efficiency, performance, and robustness to designed-
for uncertainties, i.e. ‘average’ cases. Thirdly, it conversely exhibits hypersensitivity to
unanticipated perturbations, i.e. ‘rare’ cases. This too, is a feature of most systems, as we
will see. Lastly, unlike rival generative mechanisms, the resulting structural configura-
tions are domain-specific and non-generic [9]. For a discussion of power laws, a primer
on HOT and a survey of generative mechanisms (including HOT), the reader is referred
to [10,11,12], respectively.
2.3. HOT example: Buffer overflow
We shall proceed to present a first example to highlight a HOT process-induced vulnera-

bility that can be subject to a 0th order attack.
Below we find an instantiation of a HOT model: A Probability, Loss, Resource
(PLR) optimization problem [13]; a generalized restatement of Shannon source coding
for data compression yielding the Shannon-Kolmogorov entropy for the objective func-
tion J. The reader is referred to [14] for details and more examples.
min J (1)
subject to

ri ≤ R (2)
where

J= pi li (3)
li = f (ri ) (4)
1≤i≤M (5)
We have a set of M events (Eq. 5) occurring iid with probability pi incurring loss
li (Eq. 3), the sum-product of which is our objective function to be minimized (Eq.
1). Resources ri are hedged against losses li , with normalizing f (ri ) = − log ri (Eq.
4), subject to resource bounds R (Eq. 2). We will demonstrate a mapping between this
abstracted PLR model and the following short C program (adapted from [15]) which will
be subjected to a buffer overflow.
# i n c l u d e < s t d l i b . h>
# i n c l u d e < s t d i o . h>
# i n c l u d e < s t r i n g . h>
i n t provePequalsNP ( )
{
/ ∗ Next paper . . ∗/
}
i n t bof ( )
{
ch ar b u f f e r [ 8 ] ; / ∗ an 8 b y t e c h a r a c t e r b u f f e r ∗ /
s t r c p y ( b u f f e r , g e t s ( ) ) ; / ∗ g e t i n p u t from t h e us er ∗/
/ ∗ may n o t r e t u r n i f b u f f e r o v e r f l o w e d ∗ /
return 42;
}
i n t main ( i n t a r g c , ch ar ∗∗ a r g v )
{
bof ( ) ; /∗ c a l l bof ( ) f u n c t i o n ∗/
/ ∗ e x e c u t i o n may n e v e r r e a c h
next f u n c t i o n because of overflow ∗/
provePequalsNP ( ) ;
return 1000000; / ∗ e x i t w i t h Clay p r i z e ∗/
}
We shall assume here that the probabilistic environment is adequately represented by the
user. She is asked for input in gets(), this represents the event. In the C code, the hu-
man designer specified an 8 byte buffer (char buffer[8]) and the compiler would
dutifully allocate the minimum buffer needed for 8 bytes (this is the resource r). Hence,
the constrained resources r is the variable buffer. The loss associated with the user
input event is really a step function; as long as the user satisfies the assumption of the
designer, the ‘loss’ is constant, and can be seen (simplified) as just the ‘normal’ loss in-
curred in proper continuation of control flow. Put differently, as long as user input is ≤ 8
bytes, the resource r is minimally sufficient to ensure normal control flow continuation.
If, however, the user decides to input ‘Honorificabilitudinitatibus’ (this lengthy wink to
the Bard was implicitly assumed to be an unlikely/impossible event by the human de-
signer in the code declaration), the loss l functions takes a huge step jump: a catastrophic
failure ensues since strcpy(buffer,gets()) overflows buffer. The improbable
event breaches the resource and now, control flow may be rerouted, the process crashed,
shellcode executed via a stack overflow - or in our example, fame remains elusive.
How did this vulnerability come about? In keeping with our hypothesis, we may
discern two distinct, domain-specific HOT (Highly Optimized Tolerance) optimization
processes at play - one involving human designers and the other, code compilers - that
had a hand in allocating the resource that was breached. The first domain-specific mech-
anism that induces a cost-optimized, resource-constrained structure on the executable
program is the human element. Humans using best-practice software development tech-
niques have to juggle at various stage of the design and coding stages: Evolvability vs
specificity of the system, functionality vs code size, source readability vs development
time, debugging time vs time-to-market, just to name a few conflicting objective function
and resource constraints. The second domain-specific mechanism that induces a cost-
optimized, resource-constrained structure on the executable is the compiler. The com-
piler functions as a HOT process. Cost function here include memory footprint, exe-
cution cycles, and power consumption minimization, whereas the constraints typically
involve register and cache line allocation, opcode sequence selection, number/stages of
pipelines, ALU and FPU utilization.
3. Background and Related Work
The issues of vulnerabilities in ancillary systems and their impact on end systems have
been discussed in the popular press. Makansi issues a clarion call to action - part histori-
cal, current and future US survey, part Cassandra-cry [16] - on the sorry state of the US
electricity grid. Pertinent to our discussion is his focus on the grid’s transmission subsys-
tem: Maintenance neglect of transmission lines, pylons and most importantly, the nearly-
unguarded substations. It is the opinion of the author that the neglect of the ancillary
transmission system viz. the grid system constitutes a prima facie example of constraint-
based value optimization as suggested by HOT, given that the former accounts for less
than 10% of the electricity asset value chain.
Within a more general framework of catastrophic societal scenarios, Clarke [17]
raises awareness of seldom-mentioned ancillary systems. He stresses hidden but perva-
sive technological and social interdependence and subsequently calls for a more expan-
sive definition of critical infrastructure. In the context of nth order attacks, he mentions
the essentially defenseless railway system and abounding chemical plants (a devilish tar-
get, since chemicals are very often shipped on railways through population centers). His
emphasizing near-blind spot subsystems like kindergarten teachers (in the US, around
20% of the population is in K-12 schools for about half the day) and morticians/under-
takers3 ) remains a rare and meretricious exception.4
The modeling tools provided by complex network theory have been used to evalu-
ate the susceptibility of critical infrastructure to both failure and attack. Network theory
lends itself to the main concepts of this paper, in that network graphs can be used to
represent influence diagrams, and system decomposition. In addition, through statistical
link-node distribution analysis, one is able to define a variety of centrality (vulgo ’im-
portance’) metrics (see Newman [19] for a book-length academic primer). Static social
network analysis was applied by Celebi [20, pp. 127-141] to network graphs of web-
sites affiliated with the terrorist PKK. Using graph metrics such as geodesic distance,
connectivity and principal component analysis, the goal was to identify the most influ-
ential websites (so-called hubs) order to break information connectivity; in other words,
pinpointing neuralgic nodes for removal to impede the functioning of the network.
Saddling the horse from the other end- and as a cautionary tale of what can be
learned in open societies built on trust - is the nigh unbelievable story5 of the PhD thesis
White House cybersecurity czar Richard Clarke wanted ‘burned’ in 2002. Sean Gorman,
a geography PhD student at George Mason University, gathered data on the US’s fiber
optic cable network entirely from open sources. He managed to layer the fiber-optic
infrastructure - the information backbone supporting much of the US’s military, civilian,
financial, air traffic, water, power and control critical infrastructure - onto business and
industrial sectors. The resulting map, which he could mine algorithmically with network
analysis methods for neuralgic points, was termed a ‘terrorist treasure map’. In the end,
he was allowed to publish a neutered version of his thesis [21].
From a dynamic modeling perspective in the context of TCP network/web server
request adaptation mechanisms, the paper series by Guirguis and Bestravos serve as a
good starting point [22,23]: They systematically investigate so-called Reduction of Qual-
ity (RoQ) attacks. RoQ attacks target adaptation mechanisms used in network protocols.
They achieve their effectiveness by non-DoS, low-bandwidth traffic maliciously opti-
mized against the admission controllers and load balancers, thereby continuously forc-
ing the adaptive mechanism to oscillate between over-load and under-load conditions.
Conceptually speaking, RoQ attacks may be viewed as a class of nth order attacks (1st or
2nd order degradation attacks). Fig. 4 shall help us understand the generalizable modus
operandus of RoQ attacks.
Assume the system services requests at a high steady state rate x∗ , thanks to its
adaptation subsystem that seeks to optimize service rates. Malicious traffic in form of
an RoQ attack (burst time t shaded) push the system from its steady state equilibrium;
the system, through its adaptation mechanism, slowly convergences at rate ν to the new,
lower steady state y ∗ . Since attacks have ceased, after some time, the system’s adaptation
mechanism is able to converge at a higher rate μ back to the the high steady state x∗ .
Optimized RoQ attacks would then begin anew, forcing the system to oscillate between
x∗ and y ∗ just when it has settled, thereby degrading performance of the end system.
3 From [18]: “ .. the most terrifying aspect of the epidemic was the piling up of bodies” and from historian
Alfred Crosby as quoted in [17, p.166]: “.. the accumulation of corpses will, more than anything else, sap and
even break the morale of a population”
4 Clarke’s epistemological mindset of possibilistic vs probabilistic thinking heeds poet’s William Carlos
Williams’ admonition: What would happen in a world, lit by the imagination? If on nothing else, decision
Figure 4. RoQ attacks force the adaptation mechanism with malicious traffic into dropping from a high system
steady state rate x∗ into to lower system steady state y ∗ . Picture adapted from [22, p. 3]
Putting it in the nomenclature used in this paper: The RoQ attack’s δ requests per
second for burst time t (grey shaded) repeated over period T constitutes the ‘rare event’
which the adaptation system was not expected to handle efficiently. Hence, the adaptation
mechanism - as a HOT process designed for common perturbations, but fragile towards
rare events - finds its internal assumptions (designed for normal traffic) violated. We now
move on to concrete examples of nth order attacks.
4. Example of nth order attacks
4.1. Embedding Ancillary System
Estonia, after regaining independence in 1991, decided on a massive nation-wide ‘cy-

berfication’ program: Comprehensive Internet access together with a population registry
for authentication/ID purposes would enable the Baltic nation to ‘Tiger-Leap’ into the
21th century. The result of this push was an extraordinarily far-reaching state information
system consisting of (among other things) a PKI infrastructure, over 70 state information
systems, financial institutions, state/private portals and associated data exchange layer
subcomponents.
In April/May 2007, Estonia suffered a two-phased denial of service attack (predomi-
nantly ICMP and TCP SYN6 ). The first phase (04/27/07 - 04/29/07) knocked out govern-
ment web servers and news sites, and included some semantic hacking such as web de-
makers are strongly urged to follow up on Clarke’s works.

5 See a 2003 Washington Post article at http://tinyurl.com/zuyrv
6 Nazario offers insightful traffic analysis of Estonia (http://tinyurl.com/2359fq) and the more
intense 2008 South Ossetia attack (http://tinyurl.com/6psa6r)

Attacks Destination Owner Description

35 195.80.105.107/32 pol.ee (now politsei.ee) Estonian police
7 195.80.106.72/32 www.riigikogu.ee Estonian Parliament
36 195.80.109.158/32 www.riik.ee, www.valitsus.ee State communication entry portal, Estonian Gov-
ernment
2 195.80.124.53/32 m53.envir.ee Ministry of the Environment
4 213.184.50.6/32 Estonian CERT
6 213.184.49.194/32 www.agri.ee Ministry of Agriculture
35 213.184.50.69/32 www.fin.ee Ministry of Finance
1 62.65.192.24/32 starman.ee Private telecom provider
Table 1. Second phase, 128 DDoS attacks: ICMP (115), TCP SYN (4), generic (9). Most serious 10 attacks:
10+ hours at 90 Mb/s. Peak on May 9: Attack shut down 58 sites at once. Data from Nazario (Arbor Networks)
facements. The second phase (04/30/07-05/17/07), coordinating a botnet encompassing

some 178 countries, was aimed at critical infrastructures: The two largest banks, neural-
gic routers at the ISP level and some governmental portals which were unavailable for a
couple of hours. As can be gleaned from Table 1, during the second phase of attacks, the
police, government and state communication portals, as well as the Ministry of Finance
bore the brunt of the traffic.
This case also highlights the question of perspective in classifying the level of in-
direction of an nth order attack. On one technical level, the attack could be classified as
2nd order degradation attack, since it consisted of relative primitive DoS traffic aimed
at resource allocation mechanisms underlying electronic services. From the point of the
individual, it may be classified as a 3rd or 4th order destabilization attack, since it, say,
undermined the information infrastructure needed for data exchange between the super-
market and the banks that enable him/her to use credit cards to buy groceries. For a short
description of Estonian development, a timeline of the two-phased cyber-attack that took
place and subsequent reactions, the reader is referred to [20, pp. 93-103]. We would like
to stress these cyberattacks went hand-in-hand with planned physical disruptions: SMS-
coordinated flash mobs causing traffic jams, trade and tourism interruption by train and
road blockades, physical attacks against parliament, and more. This synergistic levèe en
masse of the Russian ethnic minority to foment unrest on the ground, in conjunction with
the cyberattacks against societal critical infrastructure (see Table 1) were aimed at desta-
bilizing Estonian society. In its comprehensiveness and goals, these efforts constituted
the rare event in our model; in terms of modern conflict, it heralds a new class of ‘total
war’ (see Sec. 7).
4.2. Business Model Ancillary System
The email-born Bagle worm first appeared in January 2004 and still ranks - 5 years later
- among the top 15 malware families found in the wild, with a prevalence of roughly
2%. It reached its apex in 2006/2007, ranking among the top four, with a prevalence of
roughly 15%. For an incisive write-up, the reader is referred to [25].
What makes this worm noteworthy in our context is its 4th order attack m.o.: Through
a clever blend of so-called server-side polymorphism and ‘high variant-low instance’ re-
lease, it managed to circumvent conventional pattern-based antivirus (AV) signature de-
tection by attacking the economic cost structure of the AV companies itself. With server-
side polymorphic malware, the mutation and encryption code transform engine that pro-
duce variants is not incorporated into the individual instances, but resides remotely on a
server. This outsourcing make the job of traditional signature-based AV companies (who
Figure 5. Small batches per variant. Picture from [24].
analyze the specimens) harder, since their analysts have less of a code base to work with.
This in and of itself could have been dealt with: Bagle’s true innovation was to sabo-
tage the economic incentives of AV companies to distill such a signature by generating
enormous number of variants in very small batches.
Fig. 5 illustrates the simple but highly effective distribution approach: It lists the
average number of instances of the same code, per variant each day of the report period.
We see that very small batches of the same code were released but a huge number of
variants thereof (30’000 distinct variants server-side supplied in 2007 alone, an average
of 625 new variants a day[24]). This constitutes arguably a 4th order attack, since this
mechanism neither targeted a vulnerable program on the end system (0th order), nor dis-
abled host or server-based AV services (1st order), nor targeted (say through denial of
service or DNS rerouting) either the start or end points of the AV signature distribution
system (2nd and 3rd order), but cleverly vitiated the economic incentives of the AV com-
panies to develop signatures (4th order). With modern malware, it is simply not cost ef-
fective to invest even one day’s worth of highly skilled analyst’s time to develop signa-
tures for rapidly mutating, low-count instances - exactly the type of rare event for which
the business model was not designed.
4.3. Human Operator Ancillary System
Bond and Danezis invite the reader to entertain following Gedankenspiel [26], inspired
by Faust’s pact with Mephistoteles: Person W sends a program to person Z, accompa-
nied by an email singing said program’s praises. For it promises powers: The power to
remotely browse X’s hard disk, the power to read the emails between X and Y. Curiosity
and maybe malice piqued, Z installs the program and lo, it does not deceive: It delivers
on its promises, certainly, but surreptitiously keeps a log of Z’s activities and rummages
through Z’s files. After a critical mass of incriminating evidence is gathered, the program
now uses a combination of threats and bribes to get Z to propagate itself: From Data
Destruction (“I’ll delete all your files”) to Revelation (“I’ll tell Y you were spying on
X and Y’)’ to Reporting (“I’ll report your illegal downloads to the RIAA”) to Access
Denial (“I’ll encrypt all your files”) to Freebies (“You’ll get tons of free software”) and
the promise of more powers (“You’ll get the power to watch webcams”).
The truly devious innovation of this SATAN virus consists of very elegantly lever-
aging the psychological ancillary system of the human operator: It appeals first to a mix
of neutral (curiosity, risk) to base (greed, lust for power) psychological instincts. After a
time of reward to re-enforce the risky behavior, it then brings the full gamut of shame,
fear, cowardice and cognitive dissonance to bear in order to harness two additional sub-
systems of the human operator: His own human operator subsystem (select the next vic-
tim) and his rational subsystem (convince him/her to install me). The induced calculated
betrayal of interpersonal trust (the rare event) seems particularly odious. You can almost
see the friend exclaiming: “How could Z do this to me, as a friend?” As far as 1st or 2nd
order subversion attacks against human operators are concerned, the conceptual SATAN
virus is extraordinarily clever. 7
5. Analysis
With reference to the schematic network graph given in Fig. 6, the US national end ‘su-
per’ system of interdependent critical infrastructure ancillary systems, we outline some
characteristics for a theoretical nth order attack analysis framework.
1. We require first a notion of evolving system state, since we are dealing with dy-
namical systems.
2. Any model has to furthermore incorporate notions of cross-dependencies, since
systems are open and coupled.
3. These dependencies must include ties to assumption violations (as denoted in Eq.
2 of the HOT model in Sec. 2.2) to propagate effects between systems.
4. These propagated dependencies must have an impact on the system state that is
quantitatively measurable.
5. Lastly, the modeling formalism has to be high-level enough that there be a rea-
sonably direct correspondence between the system elements modeled and the
formalism of the approach.8
We explain the rationale for these requirements with the help of Fig. 6. For instance, the
communications infrastructure is powered primarily through the power infrastructure. If
power delivery is disrupted, telecommunications may switch to backup generators which
rely on fuel from the energy distribution infrastructure, delivered via the transportation
infrastructure paid for through the financial infrastructure. Conversely, the communica-
7 It is the author’s opinion that this conceptual SATAN virus offers one more astounding innovation, namely
symbiotic human-viral code. Even more extraordinary from the point of view of information complexity, the
probably simpler viral code manages to induce the ‘production’ of the more complex human code (propagation
module) dynamically by invoking evolutionarily and socially generated ‘factory routines’.
8 As an wished-for bonus (maybe there is a Santa Claus), model analysis should be tractable, i.e. any model-
ing approach used must try to avoid combinatorial state space explosions
Figure 6. Network of Critical Infrastructure. Picture from Sandia as shown in [27, p.12]
tions infrastructure provides control to the power and transportation infrastructure and
underlies much of the financial infrastructure. A fair question that a candidate framework
should be able to answer: How much power will we lose for how long if we degrade the
communications infrastructure’s performance by 20%?
5.1. Theoretical framework
There is a wealth of research on static network graph analysis (see [28] for a practi-
cal overview); its main drawbacks remain the inadequate handling of evolving dynamic
behavior and cross-dependencies/feedback loops. Since we are concerned with system
failure/degradation/subversion, reliability theory formalisms and models suggest them-
selves.
A first stab system decomposition into constitutive subsystems can lend itself to a
simple Fault Tree Analysis. FTA has been used for decades to model failure in multi-
component systems. Invented in 1961 by Bell Labs to improve the reliability of the Min-
uteman Launch Control System, it has since then been extensively used for evaluating
system safety in engineering disciplines as diverse as power, nuclear, electric, and source
code analysis [29]. FTA investigates independent pathways between failures of compo-
nents that lead to the fault tree’s top-event. In our parlance, this would be affecting the
the end system. Its representation takes the form of a logical diagram in which the top-
event’s occurrence depends on a specific combination of basic events, which are com-
Figure 7. A BLDMP (F , r, T, (Pi )) consists multi-top coherent fault tree F, main top event r, set of triggers
T, set of ‘triggered’ Markov processes Pi associated with leaves of F (denoted by the red dashed line), and
two categories of state for Pi , normal and failure that are triggered via appropriate transfer functions. Picture
from [31]
bined with two primary types of gates, AND and OR. Canonical FTAs have no notion
of component dependencies, or conditional event sequence timing. As such, they do not
meet our requirements; however, extensions such as ones offered in Dynamic FTA [30]
do incorporate some, but not all, requirements delineated above.
We offer one modern approach (itself a generalization of Dynamic Fault Trees) that
may be fruitfully applied for our purposes, subject to our requirements: Boolean Logic
Driven Markov Processes (BLDMP) [32]. BLDMP combines low-level global Marko-
vian state space evolution with a higher level FTA modeling approach. Each leaf is asso-
ciated with a Markov process which can model the dynamic behavior of a system. Forms
of cross-dependencies can be modeled by triggered Markov processes. This fault tree
represents the ‘structure function’ of the system. This structure imposed on the Markov
graph can be used to prune the state space, thereby avoiding combinatorial explosions
and making analysis more tractable. It remains an open question, though, whether the
quantitative impact of these propagated dependencies can be determined analytically,
given the non-linear complex dynamics of the setup, or whether one has to resort to an
mix of expert judgment, simulation results and historical empirical data.
There exists an alphabet soup’s worth of alternative formalisms describing dynam-
ical systems, each with their strength and weaknesses. We briefly mention so-called
Dynamic Reliability Block Diagrams (DRBD), as developed by [33]. In contrast to
BLMDP’s hybrid state space/combinatorial formalism, DRBD is based on the single,
high level formalism of RDB [34, Sec. 3.10]. Of interest to us is primarily its dynamic
expressiveness, which derives from a technique to model at a low level simple dependen-
cies. This basic dependency ‘building block’ can be combined with others to model any
dynamic behavior (see [35, sec. 3] ). The topic space is by no means exhausted: For an
extensive reference list of methodologies/formalisms and a reference work, the reader is
referred to [33, Sec. 6][34], respectively.
In terms of developed models, the Vulnerability of Information System (VIS) project
[36] very recently took a stab at answering questions similar to the critical infrastructure
one posed at the beginning of Sec. 5. VIS attempts to quantitatively measure the impact
of unexpected Information Communication Technology (ICT) breakdown on economic
sectors deemed potentially most ICT-vulnerable. Fondazione Formit (VIS’ project lead)
singled out five countries (Ireland, Italy, Luxemburg, Romania and Spain) and identified
key ICT-impacted economic sectors within each: Among these, we find public adminis-
tration, sewage disposal, telecommunication, finance, water and electricity supply - sec-
tors which qualify as critical infrastructure. They subsequently selected representative
companies within these sectors for micro-analysis to study ICT breakdown effects, ex-
isting recovery strategies and costs. On a macro-level (and more interesting in the con-
text of our discussion), a sophisticated econometric partial equilibrium model taking EU
sectorial interdependences as well as cascade effects into account was developed. The
model, which allowed for free variables such as ICT breakdown intensity, breakdown
and recovery time length, measured the effects of reduced ICT performance on output
and value loss, employment and price change, as well as social welfare loss with a time
horizon of one day to three months. The validity of some model output was also assessed
by means of expert judgment impact analysis in the case study companies and subsequent
country-specific micro-simulations.
As of the time of this writing (July 2009), the final report is in the works; perus-
ing available material, the numbers seem overly optimistic: Cumulative Spanish output
loss after one month (assuming 10% ICT instantaneous loss, with 50% recovery in five
days) hovers below 1% across all economic sectors. One might take issue with the strong
equilibria assumptions in the econometric model, yet the crux lies with the recovery as-
sumption: 10% ICT loss with 50% recovery within 5 days may be realistic in terms of
accidents or technological glitches, but very likely unrealizable in the face of intentional
attacks (in fairness, intentional attacks were puzzlingly not in the project scope’s risk
space). Thus we stress again, albeit in a different context, the pitfalls of strong assump-
tions, as well as the dangerous allure of fantasy recovery (‘error handling’) documents, a
topic which we shall return to below.
5.2. Practical framework
Since analytical modeling proves to be non-trivial in its requirements, perhaps an ap-

proach along the lines of a simulation offers an alternative. Indeed; given a controlled,
instrumented environment in which the end system can be situated, actual nth order at-
tacks against ancillary systems and their concomitant effects can be observed and then
evaluated. Such is the case with software application running on a single machine, where
destabilization efforts can be effected through an embedding ancillary system acting as
mediating OS middleware. We list Holodeck9; a fault injection framework that allows
Windows programs to run in simulated hostile environments [37]. Its functionality in-
cludes the ability to create resource starvation situations affecting ancillary systems such
as memory, hard disk, network bandwidth; as well as error handling ancillary system in
the form of data poisoning such as corrupted resource files/network streams, unexpected
API return values, and a gamut of explicit fault injections.
Empirical evidence collected over two decades support Holodeck’s emphasis on er-
ror handling ancillary systems. Miller subjected Unix, Windows and OS X utilities in
the simplest case to random (not malicious) keyboard input, and reported end system
crash failure rates of 25%-40%, 24%, and 7%, respectively [38][39]. Sociological and
organizational case studies by Clarke [40], analyzing what he terms ‘fantasy documents’
9 Commercially available at http://www.securityinnovation.com/holodeck/

(disaster contingency plans10 ), corroborate the brittleness of error handling subsystems,

as well.
6. Remediation
In our view, remediation efforts must either address the assumption violations underlying
the vulnerabilities, or devise a control mechanism to keep the system in a stable state,
should it come under attack. We crystallized thusly: Since we posited that the etiology
of nth order attacks (any attack) lay in the HOT-induced violations of assumptions, is
there a way of dynamically mutating those assumptions? If not, can we prevent malicious
parties from learning of these assumptions? Lastly, if we cannot prevent a violation, can
we return a system back to a stable state?
An effective, protocol compliant, but rarely used TCP feature in Linux kernels exists
which prevents some forms of degradation attacks against the TCP resource allocation
mechanism: SYN Cookies. The server outsources the state of a half-open connection
(kept normally on the server) in the form of a cryptographic challenge (the cookie) back
to the client[41]. This is an example of an assumption mutation. Internet cognoscenti
have heard of the ‘Slashdot’ effect - when legitimate connection requests overwhelm
the server because of popularity of content. This problem was tackled early on in 2001
by Akamai [42] in the form of dynamic load balancing, which constitutes a runtime
assumption mutation.
Keeping parties from learning about exact resource boundaries (and subsequent ex-
ploitation) may be able to borrow methods from thwarting so-called side channel attacks.
Side channel attacks try to infer a system process’ state by means of (sometime inadver-
tently, sometimes unavoidably) leaked observables like time to completion, EM radia-
tion, sound, protocol return values generated in course of the system’s evolution. These
attacks range the gamut from ingenious timing analysis on B-tree lookup operations and
data structure rebalancing (which lead to the release of database privileged information
[43]), to differential power analysis where current used in switching reveal activities that
can be mapped to processes [44], to CPU operation inferences through characteristic
acoustic spectral signatures [45]. In all these instances, processes leaked information. It
may be possible to design and operate systems in such a way that the leaking of resource
boundaries (the assumptions an attacker wants to violate) is minimized. We hypothe-
size that designs that incorporate the insights of Maximum Entropy Principles (for an
introduction see [46]) are a step in the right direction.
For state control, Ott’s [47] work on controlling chaotic systems may yield some
fruitful insights, since the interdependent, nested systems under consideration in this pa-
per are more than likely to exhibit non-linear, complex, chaotic behavior due to feedback
relationships. In a nutshell, Ott’s OGY method injects tiny perturbations into the system
when it threatens to veer off towards an unstable state. These perturbations (a control
vector based on the system state’s Jacobian eigenvectors) ‘nudge’ the chaotic system
back towards a fixed point and into a stable state. For a beginner’s primer on non-linear
systems, the reader is referred to [48].
10 A classic remains LILCO’s ill-fated February 13th 1986 Shoreham evacuation plan. The aim of this ex-
ercise was to demonstrate the evacuation plan feasibility. It failed at step 1: The bus drivers (a logistical and
psychological vital link; tasked among other things to evacuate children) failed [40, pp. 26-30]
7. Epilogue
In a worthwhile comparative study [49], Fukuyama of ‘End of History’ fame discusses

the notion of societal trust as a gateway to prosperity. He maintains that members of
‘high-trust’ societies (like the United States) can leverage wide-circle (beyond family
ties) trust to form efficient, optimized civic and economic organizations. It is hard to
overstate how deeply this trust subsystem permeates every facet of open societies, how
much it lowers tangible and intangible transaction costs between individuals, corpora-
tions and the state, and how easy an assumption it is to violate for malicious actors - with
disastrous effects on the end system.
This realization was not lost on Bin Laden and his fellow strategists. In a 2004 broad-
cast, he boasted (quoting research from Chatham House [50]) that the 9/11 attacks had
cost al-Qaeda only $500,000 while inflicting at least $500 billion of economic losses on
America. Accordingly, the Islamist supremacists’ playbook calls for beating the US by
systematically attacking the US economy’s vulnerabilities. The most accessible vulner-
abilities in open societies are induced by deeply ingrained trust assumptions these soci-
eties have developed over decades and take for granted: that freedom of movement, free-
dom of speech, freedom of religious assembly, assistance from the social welfare state,
immigration policy will not be used to subvert society from within; that a participant in
mass transit, a shopper at the mall, a fertilizer buyer, a student reading nuclear engineer-
ing, a worshipper at a house of prayer will not commit mass murder. The chilling passage
(excerpted from [51]) is worth quoting at length (italics are ours):
The Islamic nation has entered through al-Qa‘ida’s war with America a new period that is dif-
ferent from all the other periods experienced by Muslims against their enemies. This period
is based on economic war due to the peculiar nature of the adversary in this ferocious battle.
Usually, wars are based on military strength and victory belongs to those who are militarily
superior on the battlefield...But our war with America is fundamentally different, for the first
priority is defeating it economically. For that, anything that negatively affects its economy is
considered for us a step in the right direction on the path to victory. Military defeats do not
greatly effect how we measure total victory, but these defeats indirectly affect the economy
which can be demonstrated by the breaching of the confidence of capitalists and investors in
this nation’s ability to safeguard their various trade and dealings [..] Any operation targeting
an area of infrastructure in a new country that does not have a history of countering these op-
erations is considered as bleeding (exhausting) to the greater enemy America and the targeted
nation itself. It is so because these nations will be required to protect all similar potential
targets which results in economic exhaustion (bleeding)... For example, if a hotel that caters
to western tourists in Indonesia is targeted, the enemy will be required to protect all hotels
that cater to western tourists in all countries which may become a target of similar attacks.
You can say the same thing about living residences, economic establishments, embassies [..]
Similarly, the PRC People Liberation Army’s emphasis on asymmetric warfare and on-
going push to develop modern “Assassin’s Mace” weapons within the doctrine of “The
Inferior Defeats the Superiors”11 should give some pause. The Director of Foreign Mili-
tary Studies at the Academy of Military Sciences in Beijing, Major General Pan Junfeng,
offered following tidbits reminiscent of nth order warfare (presumably against the US) in
a 1996 issue of China Military Science (as cited in [53, p.12]):
11 Philosophic outlines of said doctrine are already found in Sun Tzu, the modern incarnation can be traced
to Mao, implementation to the 1980s, and open discussions among specialized scholars abound since the early
1990s [52]
We can make the enemy’s command centers not work by changing their data system. We can
cause the enemy’s headquarters to make incorrect judgments by sending disinformation. We
can dominate the enemy’s banking system and even its entire social order.
The interested reader is invited to peruse the some of the PLA’s official and unofficial
takes on future warfare in [54,53].
We would be remiss in our discussion if we were not to mention an nth order attack
against the ultimate ancillary system: Electromagnetic pulse attacks against the electric-
ity grid. An April 2008 report to the US House Armed Services Committee [27] outlined
the effects on critical civilian infrastructure, should a nuclear weapon12 be detonated
200-400 miles over Kansas (italics are ours):
The functioning of society and the economy is critically dependent upon the availability of
electricity. Essentially every aspect of American society requires electrical power to function.
Contemporary U.S. society is not structured, nor does it have the means, to provide for the
needs of nearly 300 million Americans without electricity. Continued electrical supply is nec-
essary for sustaining water supplies, production and distribution of food, fuel, communica-
tions, and everything else that is a part of our economy. [..] No infrastructure other than elec-
tric power has the potential for nearly complete collapse in the event of a sufficiently robust
EMP attack [..] Large-scale load losses in excess of 10 percent are likely at EMP threat levels.
Instantaneous unanticipated loss of load, by itself, can cause system collapse. This is possible
at 1 percent loss, and is very likely above 10 percent [..] Should the electrical power system
be lost for any substantial period of time, the Commission believes that the consequences are
likely to be catastrophic to civilian society. Machines will stop; transportation and commu-
nication will be severely restricted; heating, cooling, and lighting will cease; food and water
supplies will be interrupted; and many people may die.
We therefore close on a somber note: The issues touched upon in this paper are not
merely of academic or scientific interest. In practical terms, they go to the very heart of
how future conflicts between open societies and their enemies will be waged - and are
waged as we speak.
References
[1] K. Thompson, “Reflections on Trusting Trust,” CACM, vol. 27, pp. 761–764, August 1984.
[2] L. Von Bertalanffy, “An Outline of General System Theory,” British Journal for the Philosophy of
Science, pp. 134–165, 1950.
[3] P. Érdi, Complexity Explained. Springer, November 2007.
[4] G. Gigerenzer, Gut feelings: The Intelligence of the Unconscious. Viking Books, 2007.
[5] G. Torres and C. Lima, “Maximum CPU Temperature.” http://tinyurl.com/oytsnv, October 2007.
[6] US DOE, “Starlight Information System.” Pacific Northwest National Lab, 2003.
[7] G. Conti, “Attacking Information Visualization System Usability Overloading and Deceiving the Hu-
man,” in Proceedings of the 2005 Symposium on Usable Privacy and Security, pp. 89–100, ACM, 2005.
[8] W. Poundstone, Gaming the Vote: Why Elections Aren’t Fair. Hill and Wang, 2008.
[9] J. M. Carlson and J. Doyle, “Highly Optimized Tolerance: A Mechanism for Power Laws in Designed
Systems,” Physical Review E, vol. 60, no. 2, pp. 1412+, 1999.
[10] A. Clauset, C. R. Shalizi, and M. Newman, “Power-Law Distributions in Empirical Data,” SIAM Re-
views, June 2007.
[11] J. Carlson and J. Doyle, “Highly Optimized Tolerance: Robustness and Design in Complex Systems,”
Physical Review Letters, vol. 84, pp. 2529+, March 2000.
12 The exact yield necessary is presumably classified and/or undeterminable through simulation, expert
physicist estimates range from 10-1500 kilotons.

[12] M. Newman, “Power Laws, Pareto Distributions and Zipf’s Law,” Contemporary Physics, vol. 46,
pp. 323–351, September 2005.
[13] L. Manning, J. Carlson, and J. Doyle, “Highly Optimized Tolerance and Power Laws in Dense and
Sparse Resource Regimes,” Physical Review E, vol. 72, pp. 16108+, July 2005.
[14] J. Doyle and J. Carlson, “Power Laws, Highly Optimized Tolerance, and Generalized Source Coding,”
Physical Review Letters, vol. 84, p. 5656:5659, June 2000.
[15] J. C. Foster, V. Osipov, N. Bhalla, and N. Heinen, Buffer Overflow Attacks. Syngress, 2005.
[16] J. Makansi, Lights Out: The Electricity Crisis, the Global Economy, and What it Means to You. Wiley,
2007.
[17] L. Clarke, Worst Cases: Terror and Catastrophe in the Popular Imagination. University of Chicago,
2006.
[18] J. Barry, The Great Influenza: The Epic Story of the Deadliest Plague in History. Penguin, 2005.
[19] M. Newman, A.-L. Barabasi, and D. J. Watts, The Structure and Dynamics of Networks: (Princeton
Studies in Complexity). Princeton University Press, April 2006.
[20] Centre of Excellence Defence Against Terrorism, ed., Responses to Cyber Terrorism, vol. 34 of NATO
Science for Peace and Security Series E. IOS Press, 2008.
[21] S. Gorman, Networks, Security and Complexity: The Role of Public Policy in Critical Infrastructure
Protection. Edward Elgar Publishing, 2005.
[22] M. Guirguis and A. Bestavros, “Reduction of Quality (RoQ) Attacks on Internet End-Systems,” in 2005
Proceedings IEEE INFOCOM, vol. 2, March 2005.
[23] M. Guirguis and A. Bestavros, “Adversarial Exploits of End-Systems Adaptation Dynamics,” Journal
of Parallel and Distributed Computing, vol. 67, no. 3, pp. 318–335, 2007.
[24] Commtouch, “Server-Side Polymorphic Viruses Surge Past AV Defenses.” http://tinyurl.com/2vewz8,
May 2007.
[25] Commtouch, “Malware Outbreak Trend Report: Bagle-Worm.” http://tinyurl.com/39gnz4, March 2007.
[26] M. Bond and G. Danezis, “A Pact with the Devil,” in Proceedings of the 2006 Workshop on New Security
Paradigms, pp. 77–82, ACM, 2006.
[27] W. Graham, “Report of the Commission to Assess the Threat to the United States from EMP Attack:
Critical National Infrastructures,” tech. rep., Congressional Report, April 2008.
[28] W. De Nooy, Exploratory Social Network Analysis with Pajek. Cambridge University, 2004.
[29] C. Ericson, “Fault Tree Analysis – A History,” in Proceedings of the 17th International System Safety
Conference, 1999.
[30] J. Dugan and S. Bavuso, “Dynamic Fault-Tree Models for Fault-Tolerant Computer Systems,” IEEE
Transactions on Reliability, vol. 41, no. 3, pp. 363–377, 1992.
[31] M. Bouissou, “A Generalization of Dynamic Fault Trees through Boolean logic Driven Markov Pro-
cesses (BDMP),” R in Proceedings of the safety and reliability conference (ESREL07), 2007.
[32] M. Bouissou and J. Bon, “A New Formalism that Combines Advantages of Fault-Trees and Markov
Models: Boolean Logic Driven Markov Processes,” Reliability Engineering and System Safety, vol. 82,
no. 2, pp. 149–163, 2003.
[33] S. Distefano and A. Puliafito, “Dependability Evaluation with Dynamic Reliability Block Diagrams and
Dynamic Fault Trees,” IEEE Transactions on Dependable and Secure Computing, vol. 6, no. 1, pp. 4–17,
2009.
[34] M. Rausand and A. Hoyland, System Reliability Theory: Models, Statistical Methods, and Applications.
Wiley-Interscience, 2004.
[35] S. DiStefano, “How to Capture Dynamic Behaviours of Dependable Systems,” International Journal of
Parallel Emergent Distributed Systems, vol. 24, no. 2, pp. 127–150, 2009.
[36] Formit, “VIS - the Vulnerability of Information System and its Inter-Sectorial, Economic and Social
Impacts.” http://www.formit.org/vis/, May 2009.
[37] J. Whittaker, How to Break Software Security: Effective Techniques for Security Testing. Addison Wes-
ley, 2004.
[38] B. Miller, L. Fredriksen, and B. So, “An Empirical Study of the Reliability of UNIX Utilities,” CACM,
vol. 33, no. 12, pp. 32–44, 1990.
[39] B. Miller, G. Cooksey, and F. Moore, “An Empirical Study of the Robustness of MacOS Applications
Using Random Testing,” in Proceedings of the 1st International Workshop on Random Testing, pp. 46–
54, ACM, 2006.
[40] L. Clarke, Mission Improbable. University of Chicago, 1999.
[41] A. Zuquete, “Improving the Functionality of SYN Cookies,” in Proceedings of 6th IFIP Communica-
tions and Multimedia Security Conference, pp. 57–77, 2002.
[42] T. Leighton, “The Akamai Approach to Achieving Performance and Reliability on the Internet,” in
Proceedings of the 26th ACM Symposium on Principles of Distributed Computing, ACM, 2007.
[43] A. Futoransky, D. Saura, and A. Waissbein, “The ND2DB Attack: Database Content Extraction Using
Timing Attacks on the Indexing Algorithms,” in Proceedings of the 1st USENIX Workshop on Offensive
Technologies, pp. 1–9, USENIX, 2007.
[44] T. Popp, S. Mangard, and E. Oswald, “Power Analysis Attacks and Countermeasures,” IEEE Design &
Test of Computers - Design and Test of ICs for Secure Embedded Computing, vol. 24, no. 6, pp. 535–543,
2007.
[45] A. Shamir and E. Tromer, “Acoustic Cryptanalysis. On Nosy People and Noisy Machines,” tech. rep.,
RSA and MIT CSAIL, 2004.
[46] H. Kesavan and J. Kapur, “The Generalized Maximum Entropy Principle,” Systems, Man and Cybernet-
ics, IEEE Transactions on, vol. 19, pp. 1042–1052, Sep/Oct 1989.
[47] E. Ott and M. Spano, “Controlling Chaos,” Physics Today, vol. 48, no. 5, pp. 34–40, 1995.
[48] L. Lam, Nonlinear Physics for Beginners: Fractals, Chaos, Solitons, Pattern Formation, Cellular Au-
tomata and Complex Systems. World Scientific Press, 1998.
[49] F. Fukuyama, Trust: The Social Virtues and the Creation of Prosperity. Free Press, 1996.
[50] B. Hoffman and G. Weimann, “Econo-jihad,” May 2009.
[51] S. Salama, “Unraveling Al-Qa’ida’s Target Selection Calculus,” Terrorism and Political Islam, p. 41:44,
2007.
[52] J. R. Lilley and D. L. Shambaugh, eds., China’s Military Faces the Future, ch. 3-4. Studies on Contem-
porary China, M.E. Sharpe, 1999.
[53] M. Pillsbury, “China’s Military Strategy Towards the United States: A View from Open Sources,” tech.
rep., US-China Economic and Security Review Commission, November 2001.
[54] M. Pillsbury, Chinese Views of Future Warfare. National Defense University Press, September 1998.
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-282
Business and Social Evaluation of Denial

of Service Attacks in View of Scaling
Economic Counter-measures
Louis-Francois PAUa1
a
Copenhagen Business School and Rotterdam School of Management
Abstract. This paper gives an analytical method to determine the economic and
indirect implications of denial of service and distributed denial of service attacks.
It is based on time preference dynamics applied to the monetary mass for the
restoration of capabilities, on long term investments to rebuild capabilities, and of
the usability level of the capabilities after an attack. A simple illustrative example
is provided for a denial of service on a corporate data centre. The needed data
collection methodologies are categorized by classes of targets. The use of the
method is explained in the context of legal or policy driven dissuasive, retaliation
or compensation/ restoration actions. A concrete set of deployment cases in the
communications service and transport industries is discussed. The conclusion
includes policy recommendations as well as information exchange requirements.
Keywords. Cyberwar, Economics, Business impact, Social impact, Time

preference dynamics, Mobile communications, Transport , Denial of service,
Restoration costs
Introduction
This work in progress aims at addressing two strategic aspects of cyber-warfare mostly
via communications networks and IT applications: a) first to take a total economic and
social view in the assessment of evaluating damages of a cyber-warfare attacks on a
society or business target; b) scaling a trade, economic, or legal retaliation or
dissuasion for decision makers. It is assumed that the target of the attack does not in
general have itself any or sufficient defence or attack means, so that a corporate or
national level may decide ex-ante (dissuasion) or ex-post (retaliation, compensation) to
scale a business defence affecting the economic sphere of the attacker. Such an
approach is also relevant sometimes when attacker cannot be identified and localized
precisely, so that the economic sphere of the attacker is restricted to business networks
to which the attacker belongs.
Traditionally the damage assessment has been considered “binary” and limited in
time, in that the target was considered to be rendered totally dysfunctional until full
restoration only of its information and communication capabilities. Lessons learnt tell
us that other organizational, physical, human and social capabilities are to be counted
1
Correponding Author: Louis-Francois Pau, Prof. Mobile business; Email: lfp.inf@cbs.dk
L.-F. Pau / Business and Social Evaluation of Denial of Service Attacks 283
as representing often larger collateral damage of the attacks; their restoration

eventually takes quite some time, especially if the surrounding society does not have
enough civil defence means and skills in place. Vice-versa, sometimes, the
replacements made to infrastructure damaged by the attack will be less obsolete leading
to better future robustness. To address this issue, the approach is to capitalize on the
ability of cost-benefit analysis to bundle into the internal rate of return both tangible
and some intangible effects .The internal rate of return expresses the time preference on
tangible and intangible assets ,old and new, which gives a break even net present value
over the long term. It is then proposed to treat short term dynamics of this internal rate
of return , when exposed to a Brownian shock linked to an attack affecting the
command and control node for the society or business target which have their normal
long term equilibrium return rates.
Assuming the dynamic time preference resulting from a cyber-attack, it becomes
possible to estimate all of the following :a) the incremental monetary mass needed
short term for restoration of equilibrium business and social capabilities; b) long term
investment over a given pay-back horizon needed over time to restore and improve
capabilities to get back to the equilibrium rate; c) the value of the assets degraded by
the cyber-attack as short term and long term restoration measures impact the target.
Apart from relevance in a national or corporate budgeting process, such a three-
dimensional scaling of compensation, retaliation or dissuasion gives decision makers a
way to communicate efficiently around them and to implement such counter measures
against the attacker’s economic sphere while referring eventually to a game theoretical
equilibrium required by legal/treaty provisions.
As a conclusion, the proposed methodology empowers decision makers to scale
eventual economic counter-measures or threats against attackers, the efficiency of
which cannot be guaranteed as economic-social effects may not always impact
attackers but surely their surroundings, and as the resolution of decision makers may
also vary. It will be up to the reader to assess relevance in her/his own context, while
this project has assessed some concrete cases. This project has also been motivated by
specific concerns and abilities of wireless communications operators.
1. SURVEY
The cyber attacks considered in this paper (denial of service DOS , and distributed
denial of service DDOS) are those damaging information, capabilities, and sometimes
network and infrastructure elements owned or operated by a target, with resulting
damages not only to the target but also to third parties dependent on this information,
or those networks and infrastructure [1, 14, 16]. Damage assessment is considered
difficult, as the intrusions and attacks cannot always be detected short term [2, 15, 17].
Nevertheless, large economic and social impact is felt, reaching from a unit in an
organization to whole sectors; have been carried out as part of earlier work: descriptive
assessments of the impact from surveys with input-output analysis of effect from
outages and propagation models (e.g. [3, 7]), evaluations of incentives and investments
to protect the information infrastructure (e.g. [4, 6, 8]), and evaluations of cyber-
insurance premiums in relation to security procedures [5, 20]. Very few papers deal
with models for damage assessment, which would allow a company to qualitatively and
quantitatively estimate possible financial losses due to partial or complete interruption
of connectivity; in [9] a systems engineering approach is taken, while in the present one
284 L.-F. Pau / Business and Social Evaluation of Denial of Service Attacks
an economic and business approach is taken and a simple numerical example is given
in Section 4.
Also we will address in Section 5 the use of damage assessment estimates on legal
grounds for retaliation or compensation [18, 19]. A distributed Denial of Service attack
aims to deprive legitimate users of a resource or service provided by a system, by
overloading the system with a flood of data packets, thus preventing it from processing
legitimate requests. Therefore it is necessary as in [10] to take into account the
doctrines governing the allocation of liability among key players in a distributed denial
of service attack. Such doctrines are well established and based on common law tort
principles and policy considerations.
Regarding related types of attack, such as malware, viruses, identity theft,
exploiting vulnerabilities in control software / management functions/ protocols (such
as DNS and BGP errors, lack of authentification of users, services or flows, payment
solutions vulnerabilities), some studies like those of Ferris Research and Gartner
Research have shown the huge business impact thereof as well as the very high
handling plus restoration costs. But such estimates are at best interview based, and lack
an analytical framework.
2. THEORETICAL BASIS
Comparison with economic theory
One way of looking at the economic consequences of a denial-of-service is to consider

that the target has a diversity of assets included in a portfolio , each with varying life-
cycles, and that any attack affects the overall value and sustainability of the portfolio.
Whereas in economics and finance the typical research question is one of asset
allocation in view of returning some performance goals [13], the cyber-warfare
economics question is one of asset preservation over time. Another difference with
economics and finance is that in these fields’ risks and returns are usually mutualised
across populations of owners or users via legal contracts, in cyber-warfare economics
the target normally stands alone at the time of attack with all risks and must have made
all required preventive investments. The only subfield of economics where some
common features can be identified, is the area of pension economics where the retired
person wants to maintain over time a purchasing power level, although here again
assets are a mix of own assets and mutualised assets.
Regarding the definition of capabilities exposed to an attack, they are defined at
any time as the net difference between a normal time-dependent operational capability
profile of the attacked entity, and the complete or partial combined effect of the attack
and of restoration measures on normal capabilities due to the nature of the attack and
restoration processes. Consequently, dynamics play an important role, and the proposed
methodology encompasses situations with a net reduction in capabilities. If the attack
on one target involves reduced capabilities of other asset owners (like in the case of a
“netbot”, or the halving of transmission rate capability by the TCP protocol in case of a
transmission error / congestion) one can either take a systemic view or the view of the
target alone.
Regarding the description of the stochastics of attack processes, only attack
specific process specifications with related methods would allow to model them
closely, but macro-level approximations by known or tailored distributions already

provide a good basis.
Regarding the restoration process, it is also to have its specific dynamics.
However, restoration is supposed to be possible, at a cost, but not impossible, thus
implying that data protection, integrity and security must be in place. In the case of data
loss prevention (DLP) , the Ponemon Institute has estimated from commercial cases the
cost of data loss to 100 k Euro- 5000 k Euro of which 36 % due to commercial losses
and lost customers, and 36 % from loss of portable data storage. Although VOIP
content is vulnerable, repeated calls remain possible in general.
Proposed methodology: time preference dynamics
The proposed methodology is to assume that the target applies different time
preferences to the assets in its portfolio, where the time preference profiles express the
urgency at which restoration of capabilities must be carried out in view of a time
distributed attack (including a shock) degrading suddenly specific assets in the
portfolio. In economics, time preference (or "discounting") pertains to how large a
premium a user will place on usage nearer in time over more remote usage.
Taking one class of assets, assume that the time preference rate r(t) fluctuates
around an equilibrium level r(eq) while subject to a Brownian point process W(t) .The
short term dynamics are modelled by [12]:
dr(t) = a(r(eq) - r(t))dt - V.dW(t) (1)
where :
- r(t) is the short term time spot preference at time t for a given asset, 0<r(t)<1
- a is the intensity of the feedback force towards the equilibrium time preference
r(eq)
- r(eq) is the equilibrium time preference for that given asset
- V is the volatility of the time preference fluctuations
- W(t) is the stochastic Brownian point process driving the attack diffusion
process
Monetary mass requirements for restoration of capabilities
The incremental monetary mass dM (t) needed short term for restoration of equilibrium
capabilities of the asset can then be determined. Assuming for simplification purposes
the short term time preference rate r (t) to drive short term interest rate dynamics by
near a constant rM:
dM(t) = M(t-dt). (r(t)+rM).dt M(0)=M0
where :
- M(t) is the monetary mass used short term to invest in rebuilding the asset
capability to its levels just before t=0 where monetary mass represented by the
asset value was M0
- rM is the fixed increment to the short term time preference producing the short
term interest rate payable to finance the rebuild of the capabilities
Long term investments to rebuild capabilities
The long term investment K (t) over a given horizon TK needed long term to restore
and improve the assets capabilities to get back to the equilibrium time preference rate r
(eq) can be expressed as follows:
dK(t) = K(t) [r(t).dt + V.BetaK.( dW(t)+Lambda.dt) ]

BetaK = (1-exp(-a.TK))/a
where :
- K(t) is the long term bond-like investment needed over the horizon TK to
restore asset’s capabilities
- K(0) is the initial annuity value of the assets capability value over the horizon
TK
- TK is the time horizon to rebuild and possibly improve on the asset’s
capabilities; this parameter is essential in all practical cases
- Lambda is the premium by unit risk needed by the market to support the
randomness over the real time preference
- BetaK s a constant
Usability of the capabilities over time after an attack
The value of the assets degraded by the cyber-attack as short term and long term
restoration measures impact the target, is linked to a specific usability risk
characterization WA (t) of the asset’s capabilities. The change in the degree of usability
A (t) of this asset, bounded between 0 and 1 is:
dA(t) = A(t). [r(t).dt +V (dW(t)+Lambda.dt) +VA (dWA(t)+LambdA.dt)]
where:
- the first term in the parenthesis is the effect of short term restoration via the
monetary mass investment
- the second term in the parenthesis is the contribution from long term fixed
horizon asset capability rebuild
- the third term is the reduction in recovery speed linked to the volatility and
risk in the asset’s specific capabilities as they impact its degree of use
- A(t) is the effective degree of usability of the asset , A(eq)=1
- VA is the volatility of the asset’s capabilities usability risk
- WA(t) is the Brownian motion of the usability risk characterization of the
asset’s capabilities
- LambdA is the premium by time unit in unit usability risk needed by asset
users to support the randomness over the asset’s usability risk.
The unique property of this model is that all time preference variations are subject
to the short term time preference and that the risk exposure, which is here the
investment needed to restore the asset’s capabilities, is by one bond-like financing the
duration of which determines the size. The usability of the asset is a Brownian
movement correlated with the time preference rates over time. Another characteristic of
this model is that it is decoupled from the initial asset valuation , which can be tailored
to specific cases and rely on data pre-existing to an attack (see Sections 4 and 5).
3. A SIMPLE NUMERICAL EXAMPLE
Scope
This very simple example does not allow to show and exploit all the dynamic effects
taking place, but to show how a concrete situation can lead to estimations of short term
and long term financing needs tied to the time preference expressed. It also shows that,
even if financial means are made available to rebuild capabilities, the actual restoration
time of usable capabilities is very much subject to the stochastic distribution properties
and to the quality of actual means for capabilities restoration. It also leads in Section 4
to further data collection methodology considerations.
Description
The numerical example pertains to a data centre in a company, with a scrap value of 10
MEuros, running services to support company operations. The equilibrium state is one
where all services operate 100 % to support all divisions and operations with a
company turnover of 500 MEuros/year; furthermore client capabilities are dependent
on the company’s operations being supplied to them for another 500 MEuros /year
(treated as contingent liabilities). The equilibrium time preference r (eq) is equivalent
to the company’s net operational profit margin from operations r (eq) = 50 %/year,
approximated as 0.5/ (365x24) = 5.7E-05 /hour. The short term monetary interest rates
are only about 10 %/year, so that rM= -4.76E-05/hour. A full instantaneous attack W
(0) =1 on the data centre at time t=0 reduces services usability to A (0) =0 with a
minimum nominal restoration time of TK= 3 months for all resulting services and
operations to internal divisions and third parties after such a disruption. The attack lasts
dt= 1 hour , taken also as time increment, creating a shift in the time preference to a
very high spot time preference value ; the maximum which can be chosen is r(1
hour)=1, meaning the target wants perceptually all measures to be taken for immediate
recovery of the data centre . With a maximum volatility in time preference fluctuations
of V=1 /hour, the needed reactivity becomes: a ~ 5.8E-05. Post attack, the short term
time preference grows tremendously leading to a strong rise in perceived short term
monetary flows for restoration dM(1) of slightly under 10 MEuros/hour ; this
expresses the perception that the data centre must be restored at once . The total
capability value of the assets over TK=3 months is 250 MEuros with an hourly annuity
of 115 740 Euros. With a risk premium Lambda= 0.2, the initial long term investments
dK (1) needed to recoup lost supplies to customers, and to rebuild the capability, can be
estimated at about 235 M Euros. For the usability risk WA (t) a simplified linear
decreasing profile can be taken over the restoration period TK, that is WA (t) =1-
(t/TK); we also assume LambdA=0. However, the quality and efficiency of the
restoration are highly volatile especially in downstream supply chains from the
company ; this leads to the usability of the target’s capabilities only increasing again
(dA(t)/dt >0) , despite a high time preference, if the volatility VA is less than 1,2*TK .
Half of the overall capability is only restored at time 0,5/ (1,2-VA/TK) which can be
longer than TK= 3 months for some values of VA.
4. APPLICATION AREAS AND DATA COLLECTION METHODOLOGIES
This paper cannot give cases or fictive examples for all the application areas for which
economic and social impact of denial of service need to be quantified. This Section
only serves to survey such areas by categories and to give when known established
approaches to assess relevant data to be fed into the calculations.
4.1. Public services
The denial of service of public services on a national basis or on an agency basis

(administrative services, social services, water, air traffic, waste management, financial
payments), have wide ranging consequences where the indirect impact encompasses
prejudice caused to citizens (in their ability to act, to get benefits or to contribute tax
etc) measured in time lost, benefits / contributions lost, and of qualitative damage
(health, safety, administrative registrations etc). In this field, traditional cost-benefit
analysis of tangible and intangible services applies. As to the setting of the time
preference rates, they should be high for those public services where public authorities
by law have obligations of service continuity, while they would be less and derived
from minimal service obligations in other cases.
4.2. Company products and services
In this case, the applicable methodology to the data collection is the one used for
corporate liability insurance assessment. This includes loss of capabilities (physical,
raw material and service related) with their replacement, loss of revenue due to non
delivery in time, physical loss of output such as manufacturing with associated logistic
and CRM overheads , indemnification of human resources if work or life is
jeopardized, and indirect loss and damage to clients. As to the setting of time
preference rates, in-company rates should correspond to the average return on assets or
operational margin (whichever is largest) within the sector in which the company was
denied services, while the same would apply for the clients in their respective sectors.
4.3. Loss of shared infrastructure
There is no established methodology to cover loss of shared infrastructure, “critical” or

not (such as communication or transportation networks, denial of service of a satellite
by jamming, etc). However the normal approach would be to make the inventory of the
lost capabilities (physical and service related) by infrastructure operator, of lost
revenues by infrastructure operator including claims payable to customers under
contract terms, of verifiable loss and damage by individual and institutional users, and
moreover of social costs to the same. As to the setting of time preference rates, this is a
difficult issue as infrastructure suppliers quite often do not have contractual quality of
service obligations. On the contrary, suppliers of “critical” infrastructure whose control
systems may have been compromised, bear a responsibility beyond just service
provisioning, and there recovery processes may be longer. Judgment would have to be
applied to the time preference of the infrastructure operator (normally very high but not
coupled to financial rates of returns) and to the users taking diversity into account. For
users the principle of setting the time preference could be based on the tolerable
postponement of the access and use of the shared infrastructure to next normal period
(such as shift by e.g. one day, or to next available equivalent infrastructure provider).
4.4. Technology providers
Some well known technology providers in such areas as communications, software,

control systems, transport technologies, biomedical devices, etc.., may be liable to
claims by their customers for vulnerabilities in their products, although third parties are
those exploiting them. While the “customer cum users” would know the attack profiles,
while not always knowing the technical roots for the vulnerabilities, technology
providers may benefit from the proposed framework for risk assessment if they share
attack profiles with their customers. The risk assessment method in turn allows them to
quantify reasonable levels of investments in improving the technologies and their
distribution mechanisms.
5. DENIAL OF SERVICE IMPACT ANALYSIS USAGE PROCESS
The concept is to use the damage assessment methodology of Section 2 , with its
different time scales, to specific data collected by established methodologies moderated
by neutral judgement (like best practices or eventually arbitration courts) (see Section
4) , to calculate estimates of the set of damages . Such assessments must be transparent
and done by neutral parties.
The assessed damages can then be used by executive authorities for a spectrum of
actions:
Dissuasive process: preemptively to a denial of service, by policy makers or

companies, to announce that these claims would be raised if an attack occurs.
The policy makers or companies may not have evidence yet or from past cases
to identify the attackers, but may communicate to make such a categorization
of attackers credible and visible to attackers .Also, subject to proper later
judicial tracing and identification of the attackers, the policy makers or
companies would communicate that they intend to recover the amounts of the
claims by all legal means in case of an attack. As the average cost to attackers
of a cyber-attack is usually small, dissuasion followed by retaliation or
recovery may be of some concern to attackers or their backers.
Retaliation process: if the attackers are traced and identified by technical
and/or judicial means, or if strong assumptions and partial evidence exist (e.g.
from IP addresses, software code structure, software forensics, etc…), legal or
forceful retaliation would be done for the same size of claims against direct or
indirect interests of the attackers. One obvious instance of this would be to
seize quarantine or destroy the physical and communications assets used by
the attackers, or assets owned controlled by them. This may happen in a
judicial framework (with fines and penal measures) or an international treaty
framework, but may be replaced by policy maker coercitive decisions
including offensive means.
Compensation / Recovery process: if the attackers are traced and identified by
judicial means, and can be put on trial, this process would use the damage
assessments as normally done in a judicial court procedure. In this case
however the data collection methodology and data would be subject to a

contradictory evaluation, there may be issues of sovereignty leading to
inability of enforcement/ extradition, and the delays involved are normally
quite long.
“Keep silent” process: There is of course a fourth process, which is to ignore
attacks, keep silent, report nothing, and not to sue, often for “image» reasons.
It is unfortunately very common that banks, communications and
infrastructure operators so far do not report attacks and even figure out other
reasons vis-à-vis their users.
It is conjectured that the main practical relevance of the proposed method is for
dissuasive and retaliation processes, resting ultimately on the ability of the asset owner
/ target to carry out and update his own exposure valuations based on estimates related
to user and client damages (tangible and intangible).
This same conjecture is obviously reinforced by the consideration that the tracing
and identification of the true attackers may not always be possible, or may take so
much time, that the strategy to use a recovery process may not work while a dissuasive
or retaliation process may have effects when used together.
Likewise, if attackers are using innocent identifiable resources, a recovery process
would take time establishing that they are not responsible, while giving time to the
responsible attackers.
It should not be forgotten that cyber-attacks against corporate assets often are
initiated from inside the company or past employees, which too opens up for a
combination of dissuasive, retaliation and partial recovery processes.
Finally, as some types of defensive measures (such as anti-virus) have fast
deployable get-around’s known to attackers, dissuasion and retaliation processes may
in some cases be the only way forward.
6. APPLICATION CASES
This research has found its way into a number of deployment cases summarized below
spanning all categories identified in Section 4:
Public services
Case: minimal public transport service under employee strikes (Western Europe)
Contribution: the proposed method allowed determining the public damage- number of
employees on strike curve, allowing for the union and the employer to settle on a
minimum service level.
Company products and services
Case: corporate liability insurance estimate for a Scandinavian CRM provider

Contribution: the customer relationship management (CRM) company’s services were
outsourced by several operators in the communications and credit card fields. The
contracts between these operators and the CRM service supplier stipulated damage
claims should the CRM supplier not be operational. The method allowed the CRM
provider to determine the liability insurance amount it had to get cover for vis-à-vis
cyber attacks to compensate its customers.
Loss of shared communications infrastructure
Case: attack on 3G operator BSC with partial recovery via other operator(s)
Contribution: The wireless 2G and 3G base system controller manages the
connectivity to and from radio base systems (RBS). Due to bad network management
or practices, some BSC are not totally immune from certain types of attacks. When
redundancy and restoration procedures have failed, radio coverage and connectivity
may be lost unless back-up is activated from other operator’s BSC (when feasible).
Such operators have to be compensated, as well as possibly some wireless service users
under contractual terms, and total damage assessment with/without insurance may be
necessary.
Mobile networks not only provide great benefits to their users but they also
introduce inherent security issues. With respect to security, the emerging risks of denial
of service (DOS and DDOS) attacks will evolve into a critical danger as the availability
of mobile networks becomes more and more important for the modern information
society. There are ways to mitigate the attacks by adding minimal authentication to the
radio channel assignment protocol, but this too has business implications and requires
risk assessment. At the same time, via subscriber management, interoperable
management and signalling / control networks, they carry the potential for tracing and
retaliation measures, besides lawful interception in support of legal procedures. In
particular is highlighted the retaliation process which international inter-carrier
settlements allow for, as such agreements reach out worldwide.
Finally, it has been brought to the attention of the author, that other applications
exist, e.g. in the case of water distribution protection, where attacks have wide reaching
implications, and where physical-chemical forensic evidence may be collected. In this
case the attack has both time-based as well as spatial distributions.
7. CONCLUSION
While law and jurisprudence regarding denial of service and other cyber-attacks is
making slow progress in both national and international arenas, this paper presents a
quantitative approach respecting attack and restoration dynamics likely to be used in
dissuasion as well as in retaliatory processes, in the hope that ultimately attackers will
feel a largely missing retroaction. It may also allow institutions and companies to
determine by self-analysis in the presence of a given threat profile, which assets to
protect in priority on economic, business and social grounds.
In the event international organizations like GATT, European Space Policy
Institute, OECD or the European Parliament (“Declaration on the reinforcement of
international security”, 25 March 2003 and report to the Council of 11 December 2008)
also embark on putting an economic and social measure to cyber-attacks, supplemented
by constraining legal measures, instead stating of political / cultural or defence values
only, this research may give elements of the analysis.
Specific policy recommendations linked to the above research and the deployed
cases would be the following:
- in international commercial contract law, allow for compensation and

information exchange clauses whereby attackers using one party’s facilities or
services to mount an attack on the other party, may retaliate against the
attackers on the basis of damage assessment and evidence provided by the
other party; an example of this are international communications operators
inter-operator settlement procedures;
- enhance auditing procedures, to verify the basis for insurance or damage
claims in the case of cyber-attacks;
- mandate reporting and information exchange about attacks to designated
governmental bodies, for sharing of attack profiles and partial evidence (like
envisaged by the EU).
Just as technical vulnerability reduction demands collaborative efforts between
users, technology providers and operators, the business and social impact assessments
also demand such collaboration and information exchange, besides internal due
diligence. The issue is which governments, players and sectors, like the
communications industry, will take concrete steps in this direction. One reason why
this is an issue is that “patches” and additional costly imperfect technologies are too
often preferred to demanding and longer lasting technical, legal, architectural and
economic measures. It is in this context that humanities, economic and social
disciplines can clarify the way towards peace in cyber conflicts [21].
What this research does not allow to do is to account for interdependencies
between targets and attackers, or proxies to the attackers, due to cross-ownership,
exclusive agreements, shared infrastructure (buildings, communications, transport, and
energy), geo-economics and political / cultural / social influence.
REFERENCES
[1] Lech J. Janczewski, Andrew M. Colarik (Editors) (2007), Cyber Warfare and Cyber Terrorism,
[2] Boca Raton: Idea Group Inc (IGI), ISBN 1591409918, 9781591409915
[3] O. Sami Saydjari (2004), Cyber defence: art to science, Communications of the ACM,
[4] Vol 47, Issue 3 (March), 52-57
[5] Scott Dynes, Eva Andrijcic, M. Eric Johnson (2006), Costs to the U.S. Economy of Information
Infrastructure Failures: Estimates from Field Studies and Economic Data, Proceedings of the
Fifth Workshop on the Economics of Information Security, Cambridge: Cambridge University
[6] Lawrence D. Bodin, Lawrence A. Gordon , Martin P. Loeb (2005), Evaluating information
security investments using the analytic hierarchy process, Communications of the ACM, Volume
48 , Issue 2 (February) , 78 - 83
[7] Jay P. Kesan_ Rupterto P. Majuca, William J. Yurcik (2004), The Economic Case for Cyber
insurance, University of Illinois College of Law and Economics Working Papers, Paper no 2,
http://law.bepress.com/uiuclwps/papers/art2
[8] Huseyin Cavusoglu (2008), Economics of information security, in: L. Jean Camp and
Stephen Lewis (Editors), Advances in Information Security, Vol.12, Springer, US, 978-1-4020-
8089-0 (Print) 978-1-4020-8090-6 (Online)
[9] Marco Benini & Sabrina Sicaria (2008), Risk assessment in practice: A real case study,
[10] Computer communications, Vol 31, no 15, 3691-3699
[11] Bruce H. Kobayashi (2005), An Economic Analysis of the Private and Social Costs of the
Provision of Cyber security and other Public Security Goods, George Mason University School of
Law, Working Paper Series, Paper no 26, http://law.bepress.com/gmulwps/gmule/art26
[12] 9. T. Dubendorfer, A. Wagner & B. Plattner (2004), An economic damage model for large-scale
Internet attacks, in Proceedings 13th IEEE International Workshops on Enabling Technologies:
Infrastructure for Collaborative Enterprises (WET ICE) , 14-16 June, 223- 228, ISBN: 0-7695-
2183-5
[13] Meiring de Villiers (2007), Distributed Denial of Service: Law, Technology & Policy, Sydney:
University of New South Wales Faculty of Law Research Series, Paper no 3,
http://law.bepress.com/unswwps/flrps/art3
[14] Valer Bocan & Vladimir Cretu (2005), Mitigating Denial of Service Threats in GSM Networks,
The First International Conference on Availability in GSM networks, Department of Computer
Science and Engineering, Politechnica University of Timisoara, http://www.dataman.ro
[15] O. Vasicek (1977), An equilibrium characterization of the term structure, Journal of financial
economics, Vol 5, no 2, 177-188
[16] K.C. Butler, D.L. Domian (1991), Risk, diversification and investment horizon, Journal of
portfolio management, Vol 17 ,Spring, 41-47
[17] D. Ventre (2007), La guerre de l’information, Paris : Lavoisier, ISBN: 978-2-7462-1883-3
[18] J. Kizza, F. Migga Kizza (2008), Securing the information infrastructure, Cyber Tech Publishing,
ISBN: 978-1-59904-379-1
[19] T. Shimeall, P. Williams, C. Dunlevy (2001), Countering cyber-war, NATO Review, Vol 49, no
4, Winter, 16-28
[20] Jason Fritz (2008), How China will use cyber warfare to leapfrog in military competitiveness ,
Culture Mandala, Vol. 8, No. 1, October, 28-80
[21] H. Axlerod & D. Jay (1999), Crime and punishment in cyberspace: Dealing with law enforcement
and the courts, Paper presented at the SIGUCCS Conference, Denver Colorado
[22] M. Erbschloe (2001), Information warfare: How to survive cyber attacks, Berkeley, California:
Osborne/McGraw Hill.
[23] Shari Lawrence Pfleeger, Daniela Golinelli, Robin Beckman, Sarah K. Cotton, Robert H.
Anderson, Anil Bamezai, Christopher R. Corey, Megan Zander-Cotugno, John L. Adams, Ronald
Euller, Paul Steinberg, Rachel Rue, Martin C. Libicki & Michael Webber (2008), Cyber security
Economic Issues : Corporate Approaches and Challenges to Decision making, Rand Corp.
Research brief RB-9365-1
[24] M. Godelier & H. Dawod (Editors) (2004), Guerre et paix, Special issue, Thema, Paris: CNRS,
March, www.cnrs.fr/presse
IOS Press, 2009
doi:10.3233/978-1-60750-060-5-294
Virtual Plots, Real Revolution

Roelof TEMMINGHa and Kenneth GEERSb
a
CEO, Paterva, Pretoria, South Africa
b
Scientist, Cooperative Cyber Defence Centre of Excellence, Naval Criminal
Investigative Service (NCIS), Tallinn, Estonia
Abstract. It is increasingly difficult to separate ‘cyberspace’ from what we think

of as the ‘real world’. Human beings respond to stimuli from both. Threats to
persons, organizations, and governments require timely and accurate evaluation,
but cyber attackers can exploit the imperfect and maze-like architecture of the
Internet to make threat evaluation difficult. In cyberspace, it is possible to create
fraudulent online identities – potentially millions of them – that could
programmatically support any personal, political, or military agenda. In the future,
computer botnets may evolve from spam and Distributed Denial of Service
(DDoS) generators to semantic creatures that can post opinions, arguments and
threats on the Internet. Counterfeit identities on the World Wide Web (WWW),
complete with randomized or stolen biographies, pictures, and multi-year histories
of Internet activity, will be difficult to separate from real human beings because
there is no quick way to determine whether a virtual person really exists.
Keywords. Artificial Intelligence (AI), botnet, cyber threat, cyber warfare, identity
theft, Information Operations (IO)
Introduction: a ‘semantic botnet’
Cyber attackers exploit the relative anonymity of Internet communications to send

unwanted data, including spam, malicious code and Denial of Service (DoS) attacks
around the world with near impunity. In the future, computer ‘botnets’ – networks of
compromised and organized computers within a common Command and Control (C2)
infrastructure [1] – will evolve to encompass virtual populations of randomly-generated
and/or stolen identities, which could be used to support any personal, political, or
military agenda. Each fabricated virtual identity will have a ‘life’ of its own, whose
credibility grows over time as the number and variety of its Web postings proliferate.
In 1950, Alan Turing wrote that even the “dullest” human could outperform a
computer in a conversation with another human. Turing believed it was inconceivable
that a machine could provide a “meaningful answer” to a truly wide variety of
questions [2]. In 2009, that may still be the case, but there is a big difference between
the formal test that Turing proposed and posting a comment to a blog. Even with
adequate time for analysis, there is simply not enough content to evaluate whether it
was posted by a man or a machine.
On the Internet, the best way to separate real people from artificial people –
without time-consuming, in-depth and unlikely cross examination – is with statistics.
R. Temmingh and K. Geers / Virtual Plots, Real Revolution 295
However, for every mathematical defense strategy, there seems to be an effective

response for the attacker.
1. The search for intelligent life on the Web
Until recently, most Internet conversations were conducted via email, in relatively
interactive, one-on-one correspondence (Internet Relay Chat (IRC) does not count as it
was never mainstream). Today, new technologies such as YouTube, Facebook and
Twitter allow each of us to be a prolific producer of digital information. The current
model is frequently not one-to-one, but one-to-many or many-to-one.
1:n/n:1 communications are popular because they cover a lot of ground very
quickly; for example, Sophie can now update her whole family at once instead of each
member individually. However, the trade-off here is a loss of interactivity. To pass or
fail the Turing Test, some level of cross-examination is required. In short, banal, or
low-interaction conversations, a cyber attacker (in the form of an artificial, ‘proxy’
personality) will find it easier to push information to the world and have it accepted as
is, i.e. with no follow-up question and answer time.
Further, it is reasonable to assume that most Internet traffic consists of trivial,
everyday information. Serious political and philosophical discussions normally do not
take place in the 1:n/n:1 environment. Even when it does, Natural Language Processing
(computer analysis of human languages) is unproven technology, and requires
significant human oversight to be effective.
Thus, this paper assumes that most of the information found on the Internet is not
unique, and can be stolen and repackaged for nefarious purposes. Even among unique
photos of a common sight such as the Eifel Tower, the photographers themselves might
be hard-pressed to find their own picture among others. Effective authentication
technologies such as digital signatures exist, but they are rarely used for common
communications, which remain open to theft and malicious manipulation.
2. Internet-enabled intelligence
With an unrestricted Internet connection, the average Web user now has access to more
information than her head-of-state did just five years ago. All of Wikipedia, for
example, can fit on one hard drive. The data points are now all there; the best strategy
is to choose one’s sources carefully and to discriminate between good and bad analysis.
This is the art of Open Source Intelligence (OSINT).
Computer hackers conduct OSINT just like everyone else. In fact, they also begin
their search for information at a target’s homepage. Good OSINT can quickly lead an
attacker from a name to a date-of-birth, address, education, medical records, and much
more. Via social networking sites, the attacker may soon discover intimate details of a
person’s life, from the clubs she frequents to where she might physically be at any
given moment. Eventually, a web of connections to other people, places and things can
be constructed.
Good OSINT researchers – and hackers – master both the semantic side of the
Web (e.g. the content of a webpage), and the technical side, like the Domain Name
Service (DNS), or the ‘phone book’ of the Internet. The DNS registry catalogues who
owns a given website, and often provides a point-of-contact for them in the real world.
296 R. Temmingh and K. Geers / Virtual Plots, Real Revolution
Hackers ‘enumerate’, or conduct in-depth technical reconnaissance, against cyber

targets. Technical information, including barely-hidden ‘metadata’ such as an Internet
Protocol (IP) address or a timestamp, is analyzed for anything that can be exploited in
the real world. Common applications like webmail are frequently targeted. Sooner or
later, hackers normally find an open, misconfigured, or vulnerable Internet access point,
which is analogous to a thief finding or forcing open a door or window in the physical
world.
The real magic of an effective cyber attack lies in combining technical data with
real-world information. Likewise, threat actors can be divided into those who have
‘reach’ into the real world, and those who do not.
3. It’s good to be the king
The combination of Internet monotony and hacker creativity described above can make
for a volatile mixture. The average computer programmer could never pass the Turing
Test, but she can write a program that updates the world via Twitter on how a bogus
Web user is spending his day, or what a bogus Web user thinks about how you are
spending yours. And if it is theoretically possible to create one false Web identity,
perhaps millions of them already exist.
A large virtual population, scattered all over the world and encompassing different
socioeconomic backgrounds, could be programmed to support any personal, social,
business, political, military, or terrorist agenda. The nature of an attack could be limited
only by the attacker’s imagination. For example, in the week before an election, what if
both left and right-wing blogs were seeded with false but credible information about
one of the candidates? It could tip the balance in a close race to determine the winner.
Via Internet-enabled OSINT, targets can be meticulously profiled by an attacker to
learn personal, organizational, or national sensitivities and vulnerabilities. For example,
if the target were a multinational corporation (MNC) engaged in oil exploration,
OSINT might reveal a wide range of attack vectors: disgruntled employees, friction
with indigenous populations, whistleblowers, and/or ongoing lawsuits. A zombie army
could be used to target any or all of the above – including judges and jury – by
manipulating industry blogs, commenting on news articles, sending targeted email, etc.
The MNC, of course, could have its own botnet army pushing its side of the story.
In the impersonal world of cyberspace, who can say for sure whether a message
was sent by a real person? Even highly idiomatic language can be stolen by a robot and
used (perhaps incorrectly) in another context. It is beside the point to say that one could
eventually authenticate the information. Propagandists seek first and foremost to bring
attention to their cause; ethical considerations are secondary. And the attacker may
simply need for the effect to be temporary. If a certain momentum toward the desired
goal is achieved – that is, if real people begin to follow the robots – then the attacker
can begin to ‘plug out’ the artificial intelligence. The robots could then be
reprogrammed for their next assignment.
Over time, if fake users cannot be distinguished from human users on the Internet,
the latter will be forced into a situation not unlike Harrison Ford in Blade Runner. The
difference will be that there is insufficient interactivity with the robot to spot the fake.
4. The technical details
Today, botnets spam the world, perform DDoS attacks, and hack other computers.
Tomorrow, they could be used by ideologues to sway public opinion.
Programmatically, a complex, copy-and-paste algorithm can steal biographical
information from web pages, news reports, blogs, and other Internet resources. These
in turn can be reconstructed to form the skeleton of an artificial personality. Details
from popular news and current events will put meat on the bones. Once created, these
artificial ‘people’ will be instructed to begin interacting with the Web in multiple ways.
In due course, they will assume a ‘life’ of their own, and might even make a few
human friends in the process.
The following steps have been field-tested with good results:
Her name is Violet:

• Visit the Census database
(http://www.census.gov/genealogy/names/names_files.html)
• Select random first name
• Select random last name
She looks real:

• Select random but common first name/lastname combination
• Search Google (Images) for “fname lname @ Facebook” inurl:profile,
medium size with face recognition
• Select random image after page 1
She has a real job:

• Mine the LinkedIn Directory
(http://www.linkedin.com/pub/dir/fname/lname)
• Mine the ZoomInfo Developer API (http://developer.zoominfo.com)
• Pick random data and combine creatively
She said what?

• Violet opens a social networking account
• She befriends people
• She posts to their site
5. Evaluating the credibility of a cyber threat
In chess, it is often said that a threat is mightier than its execution. The very existence
of a threat – regardless of whether it can be realized – tends to have a harmful effect on
the victim, which may behave differently or even begin to act in a way that undermines
its long-term security.
OSINT can yield enough information about a target to make even an empty threat
seem credible. It is always difficult to quickly and accurately evaluate newly-
discovered information, but cyber threats are especially complicated due to the power
of modern OSINT and the relative anonymity behind which cyber attackers can hide.
For example, phishing attacks are successful even though they normally employ only
one layer of deceit: the website itself. Intelligent attackers can weave a much more
intricate web of deception than that; an entire organization could successfully be faked
if the time were taken to invest in enough third-party references.
In cyber terminology, the classic ‘I know where you live’ can be articulated as ‘I
know your Oracle server runs on 10.7.0.33, its administrator is Bob, and Bob likes
passwords that relate to Manchester United’. OSINT specialists, especially those with
some knowledge of computer hacking, could quickly develop the following threat:
‘You have an appointment today with Dr. Livingstone at the Olympic Hotel … if I
were you, I would cancel it’. Business leaders, military officers, and even heads-of-
state have personal lives that can be targeted.
Botnet armies could be used to amplify a threat or to artificially enhance its
credibility. If an attacker threatened a corporation or a government with strikes or civil
unrest, a barrage of hard-to-verify complaints on Web fora could augment the threat,
especially if the attacker had been seeding the fora for some time. The challenge for the
attacker is to make the fabricated ‘evidence’ seem real while making verification a
complex and time-consuming challenge.
When evaluating a cyber threat, it is important to remember that what makes a
cyber attack easy – the power, ubiquity, vulnerability and anonymity of the Web – can
also lessen its credibility. Good OSINT can lead to a significant bluff. In fact, the
problem of attribution is the most complicating factor in cyber threat analysis. If the
attacker is careless and leaves a large digital footprint (e.g. his home IP address), law
enforcement may be able to take quick action. If the cyber attacker is smart, and covers
his digital tracks, then deterrence, evidence collection, and prosecution become major
challenges.
In almost all cases, computer log files alone do not suffice. Unmasking a cyber
attacker requires the fusion of cyber and non-cyber data points. Investigators must enter
the real world if they want to arrest a computer hacker. There will always be clues: if
the goal is extortion, where is the money to be paid, and is there a point-of-contact? If
the threat is Denial of Service, the target could ask for a proof of capability. The point
is to generate a level of interactivity with the cyber threat actor that might be used
against it. Further, cross-checking suspect information against trusted sources is always
one of the best defenses.
From a technical perspective, solutions to the attribution problem exist. They
include the increased use of Public Key Infrastructure (PKI), Internet Protocol version
6 (IPv6), and biometrics. Neural networks have also played a considerable role in
reducing credit card fraud [3], and their ability to locate suspicious patterns in
voluminous network traffic could be helpful outside the financial sector in the future.
However, wide-scale deployment and proper implementation of such technologies are
still years away. The widespread use of anonymous email services to support criminal
activity, for example, has convinced some that an international convention is needed to
regulate its use [4].
In the short term, one inexpensive counter to the threat posed by fake online
identities is the simple use of a live video feed. As in Blade Runner, before you can
really trust someone, it may be necessary to look her in the eye.
6. Attacking zombie armies with mathematics
Cyberspace mirrors the real world, and as such, it is complex and highly dynamic.
Nonetheless, security analysts must find signals within the noise, or a targeted attack in
a sea of normal network traffic. By way of example, let us examine an attempt to hack
a simple, online poll.
The Internet Movie Database (IMDB) ranks Sergio Leone's Il buono, il brutto, il
cattivo as the top-rated ‘Western’ film of all time, with an average user-determined
score of 8.9 on a scale of 1 to 10 [5]. High IMDB rankings are lucrative in DVD sales,
so a rival production company might try to raise the value of its own, low-ranked
Western Five Bloody Graves by artificially increasing its number of high votes.
The IMDB, and the copyright holders of Il buono, must defend their turf. A sound
strategy could consist of a two-step process:
1. the discovery of statistics that distinguish humans from computer programs as

they vote in an online poll, and
2. using these statistics to support traffic analysis and database integrity.
Is it possible to separate human voters from robotic voters in a given data set? The
trick is to keep sorting the data until identifiable fault lines appear. The goal of an
attacker is to skew the poll result without being discovered; the goal of an IMDB
security analyst is to identify the artificial votes and discard them. In concrete terms,
the analyst should try to isolate portions of the data set that look different than those
created by humans. While human beings are occasionally irrational, their behavior on
the whole can be qualified and quantified as human. For example, when asked to vote
on a scale from 1 to 10, human results normally lie within a ‘bell curve’: some are high,
some low, but most votes fall somewhere in the middle.
Statistical analysis should reveal characteristics that distinguish humans from
robots throughout the entire voting process. For example, if a computer program were
to rate films in a purely random fashion, there would be no qualitative bell curve at all
(instead, an equal number of 1s, 2s, 3s, etc). In terms of voting frequency, humans may
typically cast their ballots over lunch or before bedtime; computers do not share the
same requirements for nourishment and rest, so any serious divergence on vote
frequency may be a sign of bot infestation. Humans are also prone to some highly
subjective choices: top-ranked Il buono has received over 100,000 votes, while fourth-
ranked The Wind has barely 2,000 to its credit. The Wind thus may be a ‘hidden gem’;
qualitative distinctions such as current popularity and off-beat taste may be difficult to
program accurately.
On the technical side, it is possible to analyze the Internet traffic that brought the
vote from the remote computer to IMDB in the first place. The ‘source’ Internet
Protocol (IP) address can be geo-located on the Earth with the help of DNS. A good
security analyst brings some knowledge of culture and politics to her analysis, and
understands that there should not be too large of a discrepancy between what she
expects to find and what she does find in the data.
Think of an IP address as a car. Not every parking space should be occupied by a
red, 1989 Fiat Uno, just as not every entry in a computer log file should contain the
same IP address. At the other extreme, randomizing IP addresses also does not work;
one might then see just as many Maseratis in the lot as there are Hondas. To make his
cyber attack credible, a hacker needs to make the final distribution of his source IP
addresses mirror real Internet traffic patterns, which would require a large and
sophisticated botnet.
Internet browser activity also offers computer network defenders valuable data
points for analysis. When a human accesses a webpage, she typically waits for images,
forms, and advertisements to load in the browser. Computers lack the curiosity and
patience of a human. Robotic voters may move mechanically from one data request to
the next; all such regimented Web requests should be investigated for other non-human
properties.
Finally, cyber defense against virtual army attacks should involve a statistical
analysis of the alleged identities themselves. The basic strategy is similar to a game of
‘twenty questions’. Is the user male or female? Young or old? In entertainment or
politics? Strange patterns and sudden ratio changes should be investigated. Advanced
analysis might consist of an algorithm that combines first name, last name, country of
origin, IP address and vote to known or expected baselines. Attackers can never be
completely sure of what a security analyst expects to see, so their attack will always
require some guesswork and likely entail some miscalculations.
7. Conclusion
In 2009, hackers steal data, send spam, and deny service to other computers. In the
future, they may also control virtual armies, in the form of millions of artificial
identities that could support any personal, business, political, military, or terrorist
agenda. This attack vector exists because humans now communicate via ubiquitous
software that is by nature impersonal and non-interactive. Further, given the pure
amplification power of the Internet, it is not necessary that every target fall for the scam.
And it may not matter if the ruse is eventually discovered, because the attacker may
desire to sway public opinion only for a short period of time, such as prior to an
election [6], business deal [7], or military operation [8].
Technologies exist, such as PKI, IPv6, and biometrics, to mitigate this threat.
Smart system administrators, through network traffic analysis and rigorous database
oversight, can also theoretically ensure a high level of data integrity. And if an attacker
tried to fly ‘under the radar’ by using an insignificant number of bots for an attack,
there would likely be a correspondingly insignificant impact on the target data set to
merit the effort.
Unfortunately, the widespread use of good defensive tactics and technologies is
not on the horizon. Most system administrators do not have the time, expertise, or staff
to undertake a sophisticated analysis of their own data. Furthermore, clever
programming can obfuscate many common signatures: if IP addresses and browser
settings are scattered within the attack in a realistic way, the bar for cyber defenders is
raised considerably.
For the foreseeable future, individual Web users must improve their own ability to
evaluate threats emanating from cyberspace [9]. In most cases, the key is credibility.
Illustrations from the Turing Test and Blade Runner suggest that sufficient interactivity
with a computer should reveal that it is not human. But in the 1:n/n:1 computing
environment in which we now live, the danger is that adequate dialogue is becoming
rarer all the time.
References
[1] Freiling, Felix C., Holz. Thorsten, and Wicherski, Georg. “Botnet Tracking: Exploring a Root-Cause
Methodology to Prevent Distributed Denial-of-Service Attacks”. S. De Capitani di Vimercati et al.
(Eds.): ESORICS 2005, LNCS 3679, pp. 319–335, 2005.
[2] Oppy, Graham and Dowe, David. “The Turing Test”. The Stanford Encyclopedia of Philosophy (SEP),
http://plato.stanford.edu/entries/turing-test/, 2008.
[3] Rowland, Jan B. “The role of automated detection in reducing cyber fraud.” The Journal of Equipment
Lease Financing; Spring 2002; 20, 1; pg. 2.
[4] Mostyn, Michael M. “The need for regulating anonymous remailers”. International Review of Law,
Computers & Technology; Mar 2000; 14, 1; pg. 79.
[5] Top Rated “Western” Titles, The Internet Movie Database, www.imdb.com/chart/western.
[6] Consider the enormous impact of the 2004 Madrid train bombings on Spain's national elections three
days later: “Europe: An election bombshell; Spain, a week on.” The Economist. London: Mar 20, 2004.
Vol. 370, Iss. 8367; pg. 41.
[7] Financial institutions often take the loss when their clients are defrauded: Patterson, Aubrey B. “Fighting
hackers, fraud and wrong perceptions.” American Bankers Association. ABA Banking Journal; Apr
2003; 95, 4; pg. 14. However, the court case of Ahlo, Inc. vs. Bank of America, in which malicious code
on the company's computer was likely used to steal almost $100,000 from its bank account,
demonstrated that coverage is not absolute: Cocheo, Steve. “Privacy rumblings grow louder.” American
Bankers Association. ABA Banking Journal; Jun 2005; 97, 6; pg. 56.
[8] All political and military conflicts now have a cyber dimension. The conflict between Russia and
separatists in Chechnya has clearly demonstrated the power of well-timed Internet propaganda: Geers,
Kenneth. "Cyberspace and the changing nature of warfare." SC Magazine,
http://www.scmagazineus.com/Cyberspace-and-the-changing-nature-of-warfare/article/115929/, August
27, 2008.
[9] In 2006, identity theft was already the fastest-growing crime in the United States, affecting almost 20,000
persons per day: Ramaswamy, Vinita M. “Identity-Theft Toolkit.” The CPA Journal; Oct 2006; 76, 10;
pg. 66. Nearly a third of all adults in the U.S. reported that security fears had compelled them to shop
online less or not at all: Acoca, Brigitte. “Online identity theft.” Organisation for Economic
Cooperation and Development. The OECD Observer; Jul 2008; 268; pg. 12.
IOS Press, 2009
Subject Index
agent-based modeling 88 Georgia 143, 163
ancillary system 262 global regime 106
artificial intelligence (AI) 294 highly optimized tolerance 262
assumption violation 262 historical analogies 118
automated response 249 homeland security 66
autonomic computer network identity theft 294
defence 238 information 132
Belarus 156 information operations (IO) 294
borders 88 information warfare 3
botnet(s) 163, 211, 294 intelligence monitoring 191
business impact 282 interdependent security model 88
censorship 156 Internet 156
charter ‘97 156 journalism 182
China 163 law of armed conflicts 106
computer network security 226 maneuver warfare 143
countermeasures 211 media 182
critical information infrastructure 66 media studies 182
critical infrastructure 262 mobile communications 282
cyber 132 network counter-surveillance
cyber attack(s) 66, 77, 143, 182 operation 226
cyber conflict 118 network defence 226
cyber crime 156 nth order attack 262
cyber institutional factors 18 offensive cyber capability 143
cyberpower 18 passive availability
cyber risk 66 measurement 249
cybersecurity 66, 88, 118, 182 public policy 118
cybersecurity policy 66 reinforcement learning 238
cyberspace 18, 53, 132 response evaluation 249
cyberstrategy 18 restoration costs 282
cyber threat 294 revolution in military affairs 106
cyber war(s) 3, 53, 211, 282 risk 238
cyber warfare 3, 66, 106, 143, 294 Russia 163
DDoS 163, 211 simulation 238
decision science 118 social impact 282
decision-making 118 strategy and doctrine 3
defense strategies 211 sub rosa 53
defensive cyber capability 143 terrorism 77
denial of service 156, 282 terrorist motivation 77
denial-of-service attacks 249 terrorist use of the Internet 77
domain 132 threat analysis 191
economic warfare 262 time preference dynamics 282
economics 282 transport 282
Estonia 163 zombie armies 191
IOS Press, 2009
Author Index
Allen, P.D. 132 Libicki, M.C. 53
Beaudoin, L. 238 Martini, P. 211, 249
Bilar, D. 262 Matwin, S. 238
Charvat, J.P.I.A.G. 77 Mees, W. 191
Cutts, A. 66 Moran, N. 118
Dacier, M. 191 Nazario, J. 163
Farivar, C. 182 Pau, L.-F. 282
Geers, K. v, 294 Pavlyuchenko, F. 156
Gilbert Jr., D.P. 132 Rios, B.K. 143
Hare, F. 88 Sharma, A. 3
Hughes, R. 106 Starr, S.H. 18
Jahnke, M. 249 Sulek, D. 118
Japkowicz, N. 238 Temmingh, R. 294
Klein, G. 249 Thonnard, O. 191
Knight, S. 226 Tölle, J. 249
Leblanc, S. 226 Werner, T. 211
Leder, F. 211

The Virtual Battlefield: Perspectives On Cyber Warfare

Uploaded by

Copyright:

Available Formats

The Virtual Battlefield: Perspectives On Cyber Warfare

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

The Virtual Battlefield: Perspectives On Cyber Warfare

Uploaded by

Copyright:

Available Formats

THE VIRTUAL BATTLEFIELD: PERSPECTIVES

Coordinating Series Editors: Raphael C.-W. Phan and Jianying Zhou

Vol. 2. M. Joye and G. Neven (Eds.), Identity-Based Cryptography

Amsterdam • Berlin • Tokyo • Washington, DC

Distributor in the USA and Canada

PRINTED IN THE NETHERLANDS

How can you have a security conference on “cyber warfare” when

Technical Challenges and Solutions

Research Labs, France) present ground-breaking research on the behavior of ‘zombie

proposed countermeasures on network resources. This chapter discusses GrADAR, an

About the Contributors

Christian Czosseck is a Scientist at the Cooperative Cyber Defence Centre of Ex-

mat-to-diplomat Internet communications platform deployed by the U.S. Department of

preparedness, information sharing, cyber security, and public-private partnership issues.

Part I. Strategic Viewpoints

Cyber Wars: A Paradigm Shift from Means to Ends 3

Part II. Technical Challenges and Solutions

Behavioral Analysis of Zombie Armies 191

Subject Index 303

Sun Tzu Sixth Century B.C [1]

1. Cyber Irony- The Revolution in Military Affairs

Hence if information warfare is really a revolution in military affairs then ideally it

“War is thus an act of force to compel our enemy to do our will”

These societies are socially disconnected; politically disengaged; in scientific

In current contemporary world, governments play as the political instrument in the

When the art of simultaneous parallel warfare to achieve rapid dominance is

3. Campaign planning- The orchestration of strategy

Apart from cyber reconnaissance, an important component included in the

An important part of this phase is to develop a cyber deterrence which is credible

protracted conflict. This protracted conflict is essential to gain time so as to revitalize

4. Cyber Defence – A conventional fallacy

In light of these circumstances, where the conventional dictum of cyber defence is

The strategic information warfare is capable of achieving desired strategic ends by

In order to achieve these ends, a comprehensive orchestration of strategy in the

In terms of cyber defence the conventional wisdom of treating of cyber warfare as

Towards an Evolving Theory of

Keywords: cyberspace, cyberpower, cyberstrategy, cyber institutional factors

addressed in this theory. In accord with those objectives, we present a holistic

1.1. Elements of a Theory

have generally been developed drawing on experiments and observations. Many of

Figure 1. Broad Conceptual Framework

1.5. Key Definitions

If cyberspace is a domain, it has significant practical implications. It will require

Figure 2. Cyberspace, Cyberpower, Cyberstrategy, and Institutional Factors

1.6. Required Intellectual Capital

Figure 3. Required Intellectual Capital

2.1. Theoretical Aspects of Cyberspace

2.1.1. Key Trends

Figure 4. Number of Internet Users (Millions)

Figure 5. Hard Drive Capacity

manifesting exponential improvement, making it feasible to create very portable

Figure 6. Protocol Layering and Routing Packets Across a Network

Originally, this architecture was devised by a group of colleagues for whom

effectively, without creating new security vulnerabilities? This is an on-going challenge

Figure 7. A Framework to Characterize the GIG

2.1.2. Cyberspace “Rules of Thumb” and Principles

2.2. Theoretical Aspects of Cyberpower

2.2.1. Environmental Theories of Warfare

Table 1. Military Opportunities & Risks in Cyberspace