Adversarial and Uncertain Reasoning for Adaptive Cyber Defense: Control- and Game-Theoretic Approaches to Cyber Security

270 Pages·2019·15.72 MB·English

by  Sushil Jajodia, George Cybenko, Peng Liu, Cliff Wang, Michael Wellman

t Sushil Jajodia · George Cybenko · Peng Liu ·
r
A
- Cliff Wang · Michael Wellman (Eds.)
e
h
t
-
f
o
y
-e
e
v
t
ar
u
t
SS
0 Adversarial
3
8
1 and Uncertain Reasoning
1

S
C for Adaptive Cyber Defense
N
L
Control- and Game-Theoretic Approaches
to Cyber Security
Lecture Notes in Computer Science 11830
Founding Editors
Gerhard Goos
Karlsruhe Institute of Technology, Karlsruhe, Germany
Juris Hartmanis
Cornell University, Ithaca, NY, USA
Editorial Board Members
Elisa Bertino
Purdue University, West Lafayette, IN, USA
Wen Gao
Peking University, Beijing, China
Bernhard Steffen
TU Dortmund University, Dortmund, Germany
Gerhard Woeginger
RWTH Aachen, Aachen, Germany
Moti Yung
Columbia University, New York, NY, USA
More information about this series at http://www.springer.com/series/7410
Sushil Jajodia George Cybenko
(cid:129) (cid:129)
Peng Liu Cliff Wang
(cid:129) (cid:129)
Michael Wellman (Eds.)
Adversarial
and Uncertain Reasoning
for Adaptive Cyber Defense
Control- and Game-Theoretic Approaches
to Cyber Security
123
Editors
Sushil Jajodia George Cybenko
George MasonUniversity Dartmouth College
Fairfax, VA,USA Hanover, NH,USA
PengLiu Cliff Wang
PennsylvaniaState University ArmyResearch Laboratory
University Park, PA,USA TrianglePark, NC, USA
Michael Wellman
University of Michigan
AnnArbor, MI,USA
ISSN 0302-9743 ISSN 1611-3349 (electronic)
Lecture Notesin Computer Science
ISBN 978-3-030-30718-9 ISBN978-3-030-30719-6 (eBook)
https://doi.org/10.1007/978-3-030-30719-6
LNCSSublibrary:SL4–SecurityandCryptology
©SpringerNatureSwitzerlandAG2019
Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartofthe
material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodologynow
knownorhereafterdeveloped.
Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication
doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant
protectivelawsandregulationsandthereforefreeforgeneraluse.
Thepublisher,theauthorsandtheeditorsaresafetoassumethattheadviceandinformationinthisbookare
believedtobetrueandaccurateatthedateofpublication.Neitherthepublishernortheauthorsortheeditors
give a warranty, expressed or implied, with respect to the material contained herein or for any errors or
omissionsthatmayhavebeenmade.Thepublisherremainsneutralwithregardtojurisdictionalclaimsin
publishedmapsandinstitutionalaffiliations.
Coverillustration:Figure5from‘AdaptiveCyberDefensesforBotnetDetectionandMitigation’,p.170
ThisSpringerimprintispublishedbytheregisteredcompanySpringerNatureSwitzerlandAG
Theregisteredcompanyaddressis:Gewerbestrasse11,6330Cham,Switzerland
Preface
Today’s cyber defenses are largely static. They are governed by slow deliberative
processes involving testing, security patch deployment, and human-in-the-loop
monitoring.Asaresult,adversariescansystematicallyprobetargetnetworks,pre-plan
their attacks, and ultimately persist for long times inside compromised networks and
hosts.
In response to this situation, researchers in recent years have started to investigate
variousmethodsthatmakenetworkedinformationsystemslesshomogeneousandless
predictable. Thebasicidea ofAdaptation Techniques(AT) istoengineersystems that
have homogeneous functionalities but randomized manifestations. Homogeneous
functionality allows authorized use of networks and services in predictable,
standardized ways while randomized manifestations make it difficult for attackers to
engineer exploits remotely.
Examples of AT include concepts such as Moving Target Defenses (MTD) as well
as artificial diversity and bio-inspired defenses in order to assess the extent to which
they involve system adaption for security and resiliency purposes. Unfortunately, the
majority of AT research has been focused on developing specific new techniques as
opposed to understanding their overall operational costs, when they are most useful,
and what their possible inter-relationships might be. Moreover, the AT approaches
assume stationary and stochastic, but non-adversarial, environments. Situations with
intelligent peer adversaries operating in and changing a networked environment
produce dynamic behaviors that violate these assumptions, potentially defeating these
adaptations.
This volume aims to synthesize the recent advances made by a large team of
researchers working on the same U.S. Department of Defense Multidisciplinary
University Research Initiative (MURI) project during 2013–2019.1 This project has
developed a new class of technologies called Adaptive Cyber Defense (ACD) by
building on two active but heretofore separate research areas: Adaptation Techniques
and Adversarial Reasoning.
Our research has yielded a rich repertoire of AT methods for introducing diversity
anduncertaintyintonetworks,applications,andhosts.2Moreover,wehaveinvestigated
the criteria for deciding where, when, and how to best employ available AT options.
Suchmanagementdecisionsarecomplexduetotheperformanceandsecuritytradeoffs
inherent in AT approaches. To address such challenges, this project has harnessed a
1 George Cybenko, Sushil Jajodia, Michael P. Wellman, Peng Liu, “Adversarial and uncertain
reasoning for adaptive cyber defense: Building the cyber foundation (invited paper),” Proc. 10th
Int’l.Conf.onInformationSystemsSecurity(ICISS),SpringerLectureNotesinComputerScience,
Vol.8880,AtulPrakash,RudrapatnaShyamsundar,eds.,Hyderabad,India,December2014,pages
1–8.
2 SeeChapter1forabriefsummaryofadvancesdocumentedinthisbookalongwiththepointersto
therelevantliterature.
vi Preface
broad array of Adversarial Reasoning (AR) techniques to identify effective and stable
strategies for deploying AT options in operational systems. AR combines machine
learning, behavioral science, operations research, control theory, and game theory to
addressthegoalofcomputingeffectivestrategiesindynamic,adversarialenvironments.
Thesetechniquesforceadversariestocontinuallyre-assess,re-engineer,andre-launch
their cyberattacks. By integrating game-theoretic and control-theoretic analyses for
tradeoffanalysis, ACD presentsadversaries with optimizedand dynamically changing
attacksurfacesandsystemconfigurations,therebysignificantlyincreasingtheattacker’s
workloadsanddecreasingtheirprobabilitiesofsuccess.
Thiscoherentandfocusedresearchefforthasyielded:(a)scientificandengineering
principles that enable effective ACD, and (b) prototypes and demonstrations of
technologies embodying these principlesin severalreal-worldscenarios.
Weareextremelygratefultothenumerousparticipantsfortheircontributionstothe
MURI project. In particular, it is a pleasure to acknowledge the authors for their
contributions. Special thanks go to the US Army Research Office (ARO) and Alfred
Hofmann, Vice-President of Publishing at Springer for their support of this volume
summarizing the project. We also wish to thank the Army Research Office for their
financial support under the MURI grant W911NF-13-1-0421.
August 2019 Sushil Jajodia
George Cybenko
Peng Liu
Cliff Wang
Michael Wellman
Contents
Overview of Control and Game Theory in Adaptive Cyber Defenses. . . . . . . 1
George Cybenko, Michael Wellman, Peng Liu, and Minghui Zhu
Control-Theoretic Approaches to Cyber-Security. . . . . . . . . . . . . . . . . . . . . 12
Erik Miehling, Mohammad Rasouli, and Demosthenis Teneketzis
Game Theoretic Approaches to Cyber Security: Challenges, Results,
and Open Problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
Hamidreza Tavafoghi, Yi Ouyang, Demosthenis Teneketzis,
and Michael P. Wellman
Reinforcement Learning for Adaptive Cyber Defense Against
Zero-Day Attacks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
Zhisheng Hu, Ping Chen, Minghui Zhu, and Peng Liu
Moving Target Defense Quantification. . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Massimiliano Albanese, Warren Connell, Sridhar Venkatesan,
and George Cybenko
Empirical Game-Theoretic Methods for Adaptive Cyber-Defense . . . . . . . . . 112
Michael P. Wellman, Thanh H. Nguyen, and Mason Wright
MTD Techniques for Memory Protection Against Zero-Day Attacks. . . . . . . 129
Ping Chen, Zhisheng Hu, Jun Xu, Minghui Zhu, Rob Erbacher,
Sushil Jajodia, and Peng Liu
Adaptive Cyber Defenses for Botnet Detection and Mitigation . . . . . . . . . . . 156
Massimiliano Albanese, Sushil Jajodia, Sridhar Venkatesan,
George Cybenko, and Thanh Nguyen
Optimizing Alert Data Management Processes at a Cyber Security
Operations Center. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
Rajesh Ganesan, Ankit Shah, Sushil Jajodia, and Hasan Cam
Online and Scalable Adaptive Cyber Defense. . . . . . . . . . . . . . . . . . . . . . . 232
Benjamin W. Priest, George Cybenko, Satinder Singh,
Massimiliano Albanese, and Peng Liu
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
Overview of Control and Game Theory
in Adaptive Cyber Defenses
B
George Cybenko1( ), Michael Wellman2, Peng Liu3, and Minghui Zhu3
1 Dartmouth, Hanover, USA
[email protected]
2 University of Michigan, Michigan, USA
3 Pennsylvania State University, University Park, USA
Abstract. The purpose of this chapter is to introduce cyber security
researcherstokeyconceptsinmoderncontrolandgametheorythatare
relevant to Moving Target Defenses and Adaptive Cyber Defense. We
begin by observing that there are fundamental differences between con-
trolmodelsandgamemodelsthatareimportantforsecuritypractition-
ers to understand. Those differences will be illustrated through simple
but realistic cyber operations scenarios, especially with respect to the
typesandamountsofdatarequireformodeling.Inadditiontomodeling
differences, there are a variety of ways to think about what constitutes
a “solution.” Moreover, there are significant differences in the compu-
tational and information requirements to compute solutions for various
typesofAdaptiveCyberDefenseproblems.Thismaterialispresentedin
the context of the advances documented in this book, the various chap-
ters of which describe advances made in the 2012 ARO ACD MURI.
· ·
Keywords: Control Theory Game Theory
· ·
Adaptive Cyber Defense Moving Target Defense
Autonomous Cyber Operations
1 Moving Target Defenses (MTD)
The computer systems, software applications, and network technologies that
we use today were developed in user and operator contexts that greatly val-
ued standardization, predictability, and availability. Even today, performance
and cost-effectiveness remain dominant market drivers. It is only relatively
recently that security and resilience (not to be confused with fault tolerance)
have become equally desirable properties of cyber systems. As a result, the first
generationofcybersecuritytechnologieswerelargelybasedonsystemhardening
through improved software security engineering [7,21] (to reduce vulnerabilities
and attack surfaces) and layering security through defense-in-depth [28,31] (by
adding encryption, access controls, firewalls, intrusion detection systems, and
malware scanners, for example). These security technologies sought to respect
the homogeneity, standardization, and predictability that have been so valued
by the market but at the same time increasing security.
(cid:2)c SpringerNatureSwitzerlandAG2019
S.Jajodiaetal.(Eds.):AdaptiveCyberDefense,LNCS11830,pp.1–11,2019.
https://doi.org/10.1007/978-3-030-30719-6_1
2 G. Cybenko et al.
Consequently, most of our cyber defenses remain static today. They are gov-
erned by slow and deliberative processes such as software testing [40], episodic
penetration testing [39], security patch deployment [32], and human-in-the-loop
monitoring and analysis of security events [12,24,36].
Adversariesbenefitgreatlyfromthissituationbecausetheycancontinuously
and systematically probe targeted systems with the confidence that those sys-
tems will change slowly if at all. Adversaries can afford the time to engineer
reliable exploits and pre-plan their attacks because their targets are essentially
fixedandalmostidentical.Moreover,onceanattacksucceeds,adversariespersist
for long times inside compromised networks and hosts because the hosts, net-
works, and services – largely designed for availability and homogeneity – do not
reconfigure, adapt or regenerate except in deterministic ways to support main-
tenance and uptime requirements. This creates serious information and oppor-
tunity asymmetry between IT system defenders and potential attackers [6].
Inresponsetothissituation,researchersinrecentyearshavestartedtoinves-
tigate a variety of technologies that can make networked information systems
less homogeneous and less predictable. Among the terms and concepts used to
describe such cyber defense technologies are:
– Diversity: Inspired by biological systems [23], cyber diversity is a general
concept for introducing robustness and resilience into engineered systems by
reducing common failure modes in redundant system components. That is,
the goal is to avoid technology “monocultures” [44,53]. In cyber security
systems, this is typically accomplished by introducing software or network
variants appropriately [10,16,19,30].
– Randomization: One approach to introduce cyber diversity is to randomize
specific components of an information system. Such randomization can be
done at the low level of a system’s address space to defeat certain types
of memory-based exploits [43], at the software level by generating multiple
software variants through compiler randomization [30], instruction set ran-
domization to defeat injected malware [9], or randomization of a network’s
address space [26] or protocols [33], to give just a few examples.
– Moving Target Defenses: Motivated by the observation that a moving target
is harder to hit than a fixed one, the general concept behind Moving Target
Defenses in the cyber domain is that an information system that changes
dynamically during its operation will be more difficult for an attacker to
surveil, reverse engineer and ultimately exploit with sufficient degrees of per-
sistence than a fixed target [27]. Randomization and diversity are two ways
toimplement moving targetdefensesbutnotallrandomization anddiversity
techniquesnecessarilyrealizemovingtargets.Thatisbecausesomeimplemen-
tationsofdiversityandrandomizationdonotinfactchangeduringexecution
or system recovery after an attack.
AbasicgoalofMovingTargettechniquesistoengineersystemsthathavehomo-
geneous functionalities but dynamically different manifestations. Homogeneous
functionalityallowsauthorizeduseofnetworksandservicesinpredictable,stan-
dardizedwayswhilerandomizedmanifestationsmakeitdifficultforattackersto