cyberattack detection & mitigation: a goal-oriented and

36
CYBERATTACK DETECTION & MITIGATION: A GOAL-ORIENTED AND PATTERN-BASED APPROACH by Sang-woo Moon APPROVED BY SUPERVISORY COMMITTEE: Lawrence Chung, Chair Vincent Ng Lingming Zhang

Upload: others

Post on 23-Apr-2022

6 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Cyberattack Detection & Mitigation: A Goal-Oriented and

CYBERATTACK DETECTION & MITIGATION:

A GOAL-ORIENTED AND PATTERN-BASED APPROACH

by

Sang-woo Moon

APPROVED BY SUPERVISORY COMMITTEE:

Lawrence Chung, Chair

Vincent Ng

Lingming Zhang

Page 2: Cyberattack Detection & Mitigation: A Goal-Oriented and

Copyright c© 2017

Sang-woo Moon

All rights reserved

Page 3: Cyberattack Detection & Mitigation: A Goal-Oriented and

For all who believe in the journey of education

Page 4: Cyberattack Detection & Mitigation: A Goal-Oriented and

CYBERATTACK DETECTION & MITIGATION:

A GOAL-ORIENTED AND PATTERN-BASED APPROACH

by

SANG-WOO MOON, BS

THESIS

Presented to the Faculty of

The University of Texas at Dallas

in Partial Fulfillment

of the Requirements

for the Degree of

MASTER OF SCIENCE IN

COMPUTER SCIENCE

THE UNIVERSITY OF TEXAS AT DALLAS

May 2017

Page 5: Cyberattack Detection & Mitigation: A Goal-Oriented and

ACKNOWLEDGMENTS

I thank all who made this humble journey a worthwhile adventure. I set a foot in the field

of research thanks to my supervisor, Dr. Lawrence Chung. I extend my sincere thanks to

my supervisory committee members, Dr. Vincent Ng and Dr. Lingming Zhang. Also, I

appreciate my senior in our lab, Dr. Thomas L. Hill, Eunjung Park, Sungsoo An and Haan

mo Johng and I appreciate my friends, Gon-soo Moon, Kwang hoon An and Jun-jae Lee. I

also thank the financial support/grant from the Korea Army and Korea Military Academy.

Needless to say, my foremost appreciation goes to my parents for their support and endless

love, my syster and brother.

April 2017

v

Page 6: Cyberattack Detection & Mitigation: A Goal-Oriented and

CYBERATTACK DETECTION & MITIGATION:

A GOAL-ORIENTED AND PATTERN-BASED APPROACH

Sang-woo Moon, MSCSThe University of Texas at Dallas, 2017

Supervising Professor: Lawrence Chung, Chair

Concerns for computer-related security seem real and are increasingly becoming important

just about everywhere. In particular, cyberattack — not necessarily attack through a phys-

ical means — has been drawing serious attention from the media, government, academia,

etc.

However, detecting some suspicious behavior of computer-related systems as a phenomenon

of a cyberattack has been challenging. Detection helps but should be followed by some actions

towards rectifying any undesirable behavior. A complete set of actions that can absolutely

eliminate all the undesirable behaviors seems extremely difficult, if not impossible. In this

thesis, we propose a goal-oriented and pattern-based approach to detecting and mitigating

cyber attacks. Using a pattern-based approach, knowledge and experience about similar

cyber attacks are categorized into different classes of patterns, which essentially consist of a

set of conditions for determining if a suspicious incident belongs to a particular cyber attack

pattern class and a set of actions for mitigating the cyber attack incident. Using a goal-

oriented approach, together with case-based reasoning, options are explored for detecting

and mitigating cyber attacks, tradeoffs among the options analyzed, and selection are made.

In this thesis, one application is studied for illustrating, as well as for demonstrating the

potential benefits of, our approach. The application has to do with cyber attacks from North

vi

Page 7: Cyberattack Detection & Mitigation: A Goal-Oriented and

Korea to South Korea. We feel our studies show the potential benefits for our approach — for

more precisely characterizing a cyber attack (more finer-grained) and taking more surgical

mitigating actions.

vii

Page 8: Cyberattack Detection & Mitigation: A Goal-Oriented and

TABLE OF CONTENTS

ACKNOWLEDGMENTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x

CHAPTER 1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

CHAPTER 2 RELATED WORK . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.1 Cyberattack Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.2 Case-Based Reasoning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2.3 Requirement Engineering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

CHAPTER 3 A GOAL-ORIENTED AND PATTERN-BASED APPROACH . . . . 7

3.1 Goals and Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3.2 Overall Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3.3 Preparation: Cyberattack Pattern Classification . . . . . . . . . . . . . . . . 9

3.4 Detection: Pattern Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.5 Mitigation: Rational Mitigating Action Selection . . . . . . . . . . . . . . . 14

CHAPTER 4 STUDIES: NORTH KOREA CYBERATTACK DETECTION AND MIT-IGATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4.1 Preparation Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4.2 Detection Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

4.3 Mitigation Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4.4 Observation and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4.4.1 Observation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

4.4.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

CHAPTER 5 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

5.1 Contribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

5.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

BIOGRAPHICAL SKETCH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

CURRICULUM VITAE

viii

Page 9: Cyberattack Detection & Mitigation: A Goal-Oriented and

LIST OF FIGURES

1.1 Comparison between medical clinic’s diagnosis-treatment and cyberattack detection-mitigation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2.1 AVOIDIT: A Cyber Attack Taxonomy . . . . . . . . . . . . . . . . . . . . . . . 4

2.2 Case-Based Intrusion Detection System . . . . . . . . . . . . . . . . . . . . . . . 5

2.3 A Visual Meta-Pattern for the Security Mitigation Patterns . . . . . . . . . . . 6

3.1 Cyberattack Detection & Mitigation: Overall Process . . . . . . . . . . . . . . . 8

3.2 Preparation: Cyberattack Pattern Classification . . . . . . . . . . . . . . . . . . 9

3.3 Example: North Korea Cyberattack Class . . . . . . . . . . . . . . . . . . . . . 10

3.4 Example: North Korea Cyberattack Case (Lee, Y. (2016, December)) . . . . . . 11

3.5 A Visual Meta-Pattern For Cyberattack Mitigation Pattern . . . . . . . . . . . 12

3.6 Detection: Pattern Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

4.1 Study: North Korea Cyberattack Class . . . . . . . . . . . . . . . . . . . . . . . 15

4.2 Study: Probe Class Mitigating Actions . . . . . . . . . . . . . . . . . . . . . . . 16

4.3 Study: Military Intranet Accessing Case (Lee, Y. (2016, December)) . . . . . . . 17

4.4 Study: F-15 Fighter Jet Blueprints Stealing Case (Cuthbertson, A. (2016, June)) 17

4.5 Study: 3.20. Cyberattack Case (Pearson, M (2013, March)) . . . . . . . . . . . 17

4.6 Study: July 2009 Cyberattack Case (Sudworth, J. (2009, July)) . . . . . . . . . 18

4.7 Study: Class Detecting Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 18

4.8 Study: Observation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4.9 Study: Pattern Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4.10 Study: Data’s Influence Driven Mitigation . . . . . . . . . . . . . . . . . . . . . 20

4.11 Study: Interact Task Driven Mitigation . . . . . . . . . . . . . . . . . . . . . . . 21

ix

Page 10: Cyberattack Detection & Mitigation: A Goal-Oriented and

LIST OF TABLES

3.1 Detecting Patterns from Figure 3.4 . . . . . . . . . . . . . . . . . . . . . . . . . 11

x

Page 11: Cyberattack Detection & Mitigation: A Goal-Oriented and

CHAPTER 1

INTRODUCTION

Concerns for computer-related security seem real and are increasingly becoming important

just about everywhere. In particular, cyberattack — not necessarily attack through a phys-

ical means — has been drawing serious attention from the media, government, academia,

etc.

As the importance of security is increasing, the cyberattack defense is also studied more.

But attackers (or hackers) always exploit physical (like human resources, and devices) or

non-physical (like logical) vulnerabilities by detouring defensive measures and by finding

new attack techniques. As a result, not only the number of cyberattack attempts but also

the number of ”successful” cyberattack is increased (Ponemon Institute (2016)). Like the

famous proverb, there is no perfect security. Therefore detecting and mitigating against

success cyberattack is important as well as preventing cyberattack.

However, detecting some suspicious behavior of computer-related systems as a phe-

nomenon of a cyberattack has been challenging. Detection helps but should be followed

by some actions towards rectifying any undesirable behavior. A complete set of actions

that can absolutely eliminate all the undesirable behaviors seems extremely difficult, if not

impossible.

Now, let’s think about medical clinics. When people feel the mysterious pain or strange

symptom, what should they do? They should go to doctor, and doctor will check their body

conditions, for example, blood pressure, heart rate, body temperature, X-Ray photograph,

and so on. Based on observations which doctor observed, and disease patterns which doctor

made before, he will diagnose patient. After all, doctor will say, ”Symptoms are similar to

an earlier disease pattern.” Finally, based on previous treatments, doctor will say what to

do. Dealing with cyberattack is very similar with this process (see Figure 1.1).

1

Page 12: Cyberattack Detection & Mitigation: A Goal-Oriented and

Figure 1.1. Comparison between medical clinic’s diagnosis-treatment and cyberattackdetection-mitigation

In this thesis, we propose a goal-oriented and pattern-based approach to detecting and

mitigating cyber attacks based on this analogy. Using a pattern-based approach, knowledge

and experience about similar cyber attacks are categorized into different classes of patterns,

which essentially consist of a set of conditions for determining if a suspicious incident belongs

to a particular cyber attack pattern class and a set of actions for mitigating the cyber attack

incident. Using a goal-oriented approach, together with case-based reasoning, options are

explored for detecting and mitigating cyber attacks, tradeoffs among the options analyzed,

and selection are made.

In this thesis, one application is studied for illustrating, as well as for demonstrating the

potential benefits of, our approach. The application has to do with cyber attacks from North

Korea to South Korea. We feel our studies show the potential benefits for our approach — for

more precisely characterizing a cyber attack (more finer-grained) and taking more surgical

mitigating actions.

2

Page 13: Cyberattack Detection & Mitigation: A Goal-Oriented and

CHAPTER 2

RELATED WORK

2.1 Cyberattack Classification

To select effective and surgical mitigating attacks against occurred cyberattack, we need

to detect and classify them certainly, and specifically. Subsequently, to detect and classify

cyberattacks certainly and specifically, we need appropriate method for cyberattack class

modeling. Existing researches suggest a variety of modeling method for cyberattack class.

AVOIDIOT taxonomy is the one of the latest one in this field. It suggests five features

to classify cyberattacks. Features are attack vector (AV), operational impact (OI), defense

(D), informational impact (I) and target (T). Figure 2.1 shows specific structure and values.

According to Simmons et al. (2009), this taxonomy can classify cyberattacks more specifi-

cally than existing classification models. But it still has limitations to classify cyberattack

specifically as we want. For example, AVOIDIT taxonomy cant classify cyberattack in view

of the cyberattacks access point or targeted operational organization. Existing researches

to classify cyberattacks, including AVOIDIT taxonomy, they are limited, in the numbering

characteristics, hence rather coarse-grained. Consequently, it is hard to decide on the most

specific mitigation actions.

2.2 Case-Based Reasoning

Intrusion detection system using case-based reasoning is suggested in Esmaili et al. (1996,

December) as Case-Based Intrusion Detection System (CBIDS). CBIDS is proposed to mit-

igate limitations in terms of obtaining and expressing in exsting intrusion detection system,

for example, rule-based expert intrusion detection system. The overall structure of the

CBIDS is shown in Figure 2.2 . The system using audit records (after audit translating

phase) as input to the IDS module, and they will be classified into similar case in case-based

3

Page 14: Cyberattack Detection & Mitigation: A Goal-Oriented and

Figure 2.1. AVOIDIT: A Cyber Attack Taxonomy

4

Page 15: Cyberattack Detection & Mitigation: A Goal-Oriented and

reasoning module. And actions will be taken from Action table module, based on the sim-

ilar case. CBIDS helps find some common elements about general intrusion detection, and

geneal mitigating actions based on certain cases. But, when CBIDS acquires knowledge (or

information) from audit records, it doesnt classify case into class. It just classifies cases into

similar case. So, sometimes it is hard to decide adequate mitigation actions for cyberattacks.

Figure 2.2. Case-Based Intrusion Detection System

2.3 Requirement Engineering

After we detect cyberattack correctly, certainly, and specifically, some problems are still ex-

isting in terms of making cyberattack mitigating actions correctly and broadly, and selecting

several appropriate, surgical actions among them. The difficulties come from that there are

5

Page 16: Cyberattack Detection & Mitigation: A Goal-Oriented and

many limitations to have enough background knowledge of mitigating actions, especially

regarding to specific application and organization.

One of the research paper on this difficulty is Supakkul et al. (2009, August). It sug-

gest ”A Visual Meta-Pattern for the Security Mitigation Patterns” (see Figure 2.3). To

make/explore mitigating actions, it captures three kind of problems, undesirable outcome,

threat, and vulnerability. Alternative mitigating solutions/actions are explored in terms of

each problem. The optional mitigating actions are identified with relationshiop on other

non-functional requirements. These requirements are used as criterion when we decide our

selection which we take into actions. It suggests some appropriate strategies for some general

types of security vulnerability mitigation. But It focuses on proactive/preventive mitigation

patterns about general security vulnerabilities, without concern for reactive/curative cyber-

attack mitigation patterns.

Figure 2.3. A Visual Meta-Pattern for the Security Mitigation Patterns

6

Page 17: Cyberattack Detection & Mitigation: A Goal-Oriented and

CHAPTER 3

A GOAL-ORIENTED AND PATTERN-BASED APPROACH

3.1 Goals and Problems

There are a variety of researches on the cyberattack detection and mitigation strategy. But,

as mentioned earlier, existing works have weakness partially. Synthetically, we materialize

the problems like below. During detecting cyberattack, existing cyberattack detection mech-

anisms are coarse-grained, and even then, its difficult to find information about them. As

we discussed earlier, former mechanisms, for example, AVOIDIT cyberattack taxonomy and

Case-Based Intrusion Detection System can be examples. Consequently, during mitigating

cyberattack, some actions which are taken by stakeholders can be overly general, unneces-

sary, or even harmful. So, we need a better way to more precisely characterize a cyber attack

(more finer-grained) and take more surgical mitigating actions.

3.2 Overall Process

In this paper, to solve problems which are mentioned previous section, we propose A Goal-

Oriented and Pattern-Based Approach to Detect and mitigate Cyberattack. Our method-

ology consists of three phases, preparation, detection, mitigation phase. Preparation phase

is needed to set up knowledge base of our methodology. When we observed strange sighs,

we try to find suspicious cyberattack during detection phase. After find suspicious cyber-

attack, we will select appropriate mitigation actions among alternatives. Overall process is

illustrated in Figure 3.1.

7

Page 18: Cyberattack Detection & Mitigation: A Goal-Oriented and

Figure 3.1. Cyberattack Detection & Mitigation: Overall Process

First, during preparation phase, our goal is building cyberattack pattern classes. In this

phase, we will construct cyberattack classes by using some domain knowledge (for example,

general cyberattack taxonomy, suspicious attackers, and applying field). Next, for each

class, we will find detecting conditions by classifying, initiatingand patterning cyberattack

cases which happend before. And then, for each class, we will explore cyberattack mitigating

actions by using a goal-oriented approach. Second, during detection phase, our goal is finding

most suspicious cyberattack pattern class. Based on our observation, find most pattern-

matched cyberattack class. And we call it as most suspicious cyberattack class. Thrid,

during mitigation phase, our goal is selecting most effective, surgical mitigating actions on

most suspicious cyberattack pattern class. On a goal-oriented approach, we need to consider

goals. Based on most susficious cyberattack class’s mitigating actions, and the goal which

the stakeholders pursue, we will select more effective and surgical mitigating actions and put

them into action.

8

Page 19: Cyberattack Detection & Mitigation: A Goal-Oriented and

3.3 Preparation: Cyberattack Pattern Classification

The goal that we want to achieve during preparation phase is building cyberattack pattern

classes. After this phase, we will have the cyberattack pattern classes including detecting

conditions and mitigating actions for each classes. The overall process is Figure 3.2.

Figure 3.2. Preparation: Cyberattack Pattern Classification

First, we will build cyberattack classes without detecting conditions and mitigating pat-

terns. Cyberattack pattern classes are classification structures to distinguish cyberattacks.

In very general, cyberattack can be classified based on latest research like AVOIDIT a cy-

berattack taxonomy (Simmons et al. (2009)). Although this taxonomy classifies cyberattack

based on diverse characters (attack vector, operational impact, defense, informational im-

pact, and target), but it doesn’t reflect the specialities on the specific domain (for example,

characters of specific attackers and attacked organizations). And also it is hard to extend to

finer-grained cyberattack classes because it is static. So, to build our cyberattack class model,

we will use domain knowledge as much as we want. Figure 3.3 shows one of the example cy-

9

Page 20: Cyberattack Detection & Mitigation: A Goal-Oriented and

berattack classes when we use characters, NorthKorea’s Cyberattack, PhysicalNon-physical

and Operational Impacts. We can make general classes by using existing standard like in-

formation impact, and we can also make more ”finer-grained” classes based on the adjusted

field’s characters.

Figure 3.3. Example: North Korea Cyberattack Class

Next, we will find detecting conditions of each class by using existing cyberattack cases.

Figure 3.4 is the one of the example of North Korea cyberattack. Actually, in the majority of

cases, causes of cyberattack cases are in behind of veil before we spent a lot of times to inves-

tigate it. Because it is hard to exploit from well-known vulnerabilities, by using well-known

skills, and techniques, attackers usually use un-known vulnerabilities, skills, techniques, and

so on. Thus, as we can see in the article above, there are few informations about cyberattack

case. Now what we will do is finding any informations from that case and make them into

detecting patterns. In this case, we can find these patterns (see Table 3.1).

10

Page 21: Cyberattack Detection & Mitigation: A Goal-Oriented and

Figure 3.4. Example: North Korea Cyberattack Case (Lee, Y. (2016, December))

Table 3.1. Detecting Patterns from Figure 3.4

Pattern Value NoteSeverity Unknown -

Organization Government Defense MinistryNetwork Type Intranet, Internet -

Date 2017, August -Access Point Shenyang, China Key operation points for N.Korea hackers

Similar Malware Used N.Korea -

Identified patterns of this case (for example, organization, network type, access point, and

so on) are detecting conditions of classified cyberattack pattern class. This case is classified

into <Probe isA Non-physical isA North Korea Cyberattack>, and each value become de-

tecting condition of this class. Next step, we will make miticating actions for each class based

on a goal-oriented approach. One of the important point is that we will focus on reactive mit-

igating actions. Because, there are a lot of researches which are focused on preventive actions,

11

Page 22: Cyberattack Detection & Mitigation: A Goal-Oriented and

Figure 3.5. A Visual Meta-Pattern For Cyberattack Mitigation Pattern

including Supakkul et al. (2009, August). But in real world, when we detect the cyberattack

which infiltrates through proactive/preventive defense, we need to focus on reactive/curative

mitigating actions. In the result, our methodology is focused on mitigating cyberattack af-

ter detecting it, we will suggest A Visual Meta-Pattern for Cyberattack Mitigation Pattern

which is focused on reacitve mitigating actions (see Figure 3.5). Figure 3.5 is derived from

Supakkul et al. (2009, August). I will simply explain Figure 3.5, if you want to study further,

look Supakkul et al. (2009, August). Figure 3.5 consists of three parts, a context, a problem,

and its solutions. There are cyber security context which is located on the outer area (left,

right, top side on graph). Problems consist of undesirable outcome, threat, and vulnerability.

And there are mitigating actions related to problems. Each problems/mitigating actions has

contribution to context. The meaningful difference between our visual meta-pattern and Su-

pakkul et al. (2009, August)’s visual meta-pattern is solution part. Unlikely existing one, our

visual meta-pattern is focused on recovery/control/contain/prevent-re-use mitigating actions

against undesirable outcome, threat and vulnerability. By using a visual meta-pattern for

cyberattack mitigation pattern, we can broadly and deeply explore on specific cyberattack

mitigating actions with considering valuable goals.

12

Page 23: Cyberattack Detection & Mitigation: A Goal-Oriented and

3.4 Detection: Pattern Matching

In detection phase, our goal is finding most suspicious cyberattack class based on observations

and detecting conditions of each cyberattack pattern classes. We will find most suspicious

cyberattack class by using pattern-matching. In other words, we defind most suspicious

cyberattack class is most pattern-matched one compared to observations. In this paper, we

suggest the one of the simplest way to do pattern-matching process. We give values to each

pattern as the one of <Matched, Similar, Unmatched, No information>. <Matched> means

that in terms of this pattern, observation and this case totally agree. <Similar> means that

both don’t agree but almost similar, and <Unmatched> means that totally different, <No

information> means that one of them has no value on this pattern. First, we try to find

the case which has highest number of <Matched> pattern. If there are cases which have

same number of it, next we search the case which has highest number of <Similar> valued

pattern, and if some cases have same number again, try to find the case which has lowest

number of <Unmatched> pattern. If there are same number of pattern again, we treat

them as same degree of similarity. For example, in Figure 3.6, Probe pattern class has

more matched detecting condition patterns than DoS pattern class. In this situation, most

suspicious cyberattack class (as known as most pattern-matched cyberattack class) is Probe

pattern class.

13

Page 24: Cyberattack Detection & Mitigation: A Goal-Oriented and

Figure 3.6. Detection: Pattern Matching

3.5 Mitigation: Rational Mitigating Action Selection

When we reached this phase, we already detected most suspicious cyberattack basd on

cyberattack pattern classes and observations. In this phase, stakeholders can prioritize

and select cyberattack mitigation actions based on explored mitigation actions which is

made during preparation phase, detected during detection phase. To prioritize, stakeholders

should decide which goal they pursue, and based on that goal, they can check whole explored

mitigation actions and compare each alternatives, and select some of them. I will show this

process in the study.

14

Page 25: Cyberattack Detection & Mitigation: A Goal-Oriented and

CHAPTER 4

STUDIES: NORTH KOREA CYBERATTACK DETECTION AND

MITIGATION

4.1 Preparation Phase

In this chapter, we will apply our overall methodology in specific subject, North Korea

Cyberattack. As we mentioned earlier, our methodology consists of three phases, and starts

with preparation phase. First, during preparation phase, we will classify cyberattack patterns

with detecting conditions and mitigating actions. First of all, we should make cyberattack

class. To simplify our work, we will use North Korea cyberattack class as we mentioned

earlier in Section 3.3 (see Figure 4.1).

Figure 4.1. Study: North Korea Cyberattack Class

Next, we will make mitigation pattern by using a visual meta-pattern for cyberattack

mitigation pattern. Figure 4.2 is example of mitigating actions which is related to Probe

class. Probe cyberattack’s undesirable outcome is disclosure of data. Threat of this unde-

sirable outcome is Probe cyberattack on intranet, and this threat can be decomposed into

malicious use on intranet and external data transfer from intranet. This external data tranfer

from intranet is exploit vulnerability like external access on intranet.

15

Page 26: Cyberattack Detection & Mitigation: A Goal-Oriented and

To control the impact of this undesirable, we can change value to prevent attacker from

taking additional benefits by using data. If influence of data is strong, this mitigating

action is useful. And to prevent re-use the vulnerability, external access on intranet, we can

explore isolation of affected hosts or fully prohibition of external access. Former one hurts

the influence of data, but it some helps importance of interact task. And later one makes

influence of data, but it break importance of interact task.

Figure 4.2. Study: Probe Class Mitigating Actions

Next, we will investigate and classify some cyberattack cases. I will investigate North

Korea’s cyberattack cases (Figure 4.3, Figure 4.4, Figure 4.5, Figure 4.6) based on each

internet news. Each case is already classified into cyberattack class, therefore we can find

detecting conditions of classes (Figure 4.7) where those cases belong to.

16

Page 27: Cyberattack Detection & Mitigation: A Goal-Oriented and

Figure 4.3. Study: Military Intranet Accessing Case (Lee, Y. (2016, December))

Figure 4.4. Study: F-15 Fighter Jet Blueprints Stealing Case (Cuthbertson, A. (2016, June))

Figure 4.5. Study: 3.20. Cyberattack Case (Pearson, M (2013, March))

17

Page 28: Cyberattack Detection & Mitigation: A Goal-Oriented and

Figure 4.6. Study: July 2009 Cyberattack Case (Sudworth, J. (2009, July))

Figure 4.7. Study: Class Detecting Conditions

4.2 Detection Phase

As a result of the process so far, we finally have North Korea cyberattack pattern class

including detecting conditions and mitigating actions. Now, we assume that we observe

strange signs on our system. Assumed observation is Figure 4.8. This observation consists

of detecting patterns, suspect, information impact, severity and so on. During detection

phase, what we should do with this obervation is finding most suspicious cyberattack class.

In this study, we will compare Probe class and DoS class with the observation. As we can

18

Page 29: Cyberattack Detection & Mitigation: A Goal-Oriented and

see on Figure 4.9, Probe class has six matched conditions, DoS class has three matched

conditions. In other words, Probe class has more matched detecting conditions with the

observation. Consequently, Probe class is the most pattern-matched, and most suspicious

cyberattack class based on the observation.

Figure 4.8. Study: Observation

Figure 4.9. Study: Pattern Matching

4.3 Mitigation Phase

Lastly, we need to select appropriate, surgical mitigating actions among explored mitigating

actions. In this study, during detection phase, we found that Probe class is the most suspi-

19

Page 30: Cyberattack Detection & Mitigation: A Goal-Oriented and

cious cyberattack class. Now we will demonstrate how to select mitigating actions based on

Probe class’s mitigating actions and the goal which stakeholders pursue.

First, assume that stakeholders pursue data’s influence rather than importance of interact

task. Pursuing it means that stakeholders give more value on it than other one. In this

case, stakeholders select ”change data’s value to prevent attacker from taking additional

benefits” and ”fully prohibition of external access” because they help the data’s influence

goal, so this selection satisfies stakeholders purpose (see Figure 4.10). Second, on other

way, if stakeholders pursue importance of interact task, they will select other alternative

mitigiating action, ”isolation of affected hosts”. Because former actions break or hurt this

goal, but later action helps this goal. We can check this relationship on Figure 4.11.

Figure 4.10. Study: Data’s Influence Driven Mitigation

20

Page 31: Cyberattack Detection & Mitigation: A Goal-Oriented and

Figure 4.11. Study: Interact Task Driven Mitigation

4.4 Observation and Discussion

4.4.1 Observation

The two studies show both similarities and differences. The same mechanism is applied to

recognize a set of user observations as an instance of an existing cyber attack pattern class

and then to determine what kind of mitigating actions should be taken.

Due to the differences in the number and nature of detection conditions, including the

specificity, the effort and time taken to detect a cyber attack incident as an instance of a

cyber attack pattern class seem to vary.

Regardless of such differences, however, we feel that our proposed framework helps detect

a suspicious incident as a particular kind of cyber attack and decide what mitigating actions

to take. Our goal-oriented approach helps with exploration of what kinds of cyber attack a

suspicious behavior might be about. Our pattern-based approach helps organize and utilize

knowledge and experience about cyber attack detection and mitigation.

21

Page 32: Cyberattack Detection & Mitigation: A Goal-Oriented and

Perhaps most importantly, at least these studies, albeit limited in volume and variety,

show the potential for more precisely characterizing a cyber attack (more finer-grained) and

taking more surgical mitigating actions.

4.4.2 Discussion

We feel there are several fronts of uncertainties with our studies.

Due to the limited number and types of studies we have carried out, how generally and

broadly our proposed framework can help awaits more and a variety of studies. For one

thing, in addition to some focus on cyber attacks concerning national security, in particular,

North Korean cyber attack, more commercially-oriented cyber attacks would need to be

studied.

Related to the above point, a more and diversified cyber attack instances should be

mined, e.g., from literature or newspaper articles, into forming classes. Then, classes should

be related, through classification, association, dependency, etc.

For the studies, we took somewhat simplying cases, where there is an exact match between

the symptoms of an incidence that users observe and the characteristics of some pattern class.

This, of course, is likely to be somewhat far-fetched. For example, information about some

detection conditions, which are defined in a cyber attack pattern class, may not be available,

due to various reasons. For another example, a spefic user observation may be close but

not quite the same as defined in any cyber attack pattern classes. We would need to the

notions of similarity between the symptoms of an incident and the definitions of the detection

conditions. Furthermore, it would also be helpful to bring in the notion of weight on different

detection conditions, since they may have different criticality values.

We also felt that manual detection and mitigation seems time-consuming and difficult.

22

Page 33: Cyberattack Detection & Mitigation: A Goal-Oriented and

CHAPTER 5

CONCLUSION

5.1 Contribution

This overall work will help stakeholders detect a suspicious incident as a cyber attack case

based on the pattern-based appoach and decide what mitigating action to take based on the

goal-oriented approach. Pattern-based approach helps organize and utilize knowledge and

experience about cyber attack detection and mitigation. Goal-oriented approach helps with

exploration of what kinds of cyber attack a suspicious behavior might be about. Studies

show the potential for more precisely characterizing a cyber attack (more finer-grained) and

taking more surgical mitigating actions.

5.2 Future Work

The accuracy and the possibility of detecting depend greatly on classes and patterns we

made. We know how much we can see, and we see how much we know. So, more pattern

classes have to be conducted in a study. And in this methodology, because different fields

have each characters, more studies in a variety of applications are needed. And also there

are needs for decrease time during each phase, and software support tools for each phases

are needed.

23

Page 34: Cyberattack Detection & Mitigation: A Goal-Oriented and

REFERENCES

Cuthbertson, A. (2016, June). NORTH KOREAN HACKERS STEAL F-15 FIGHTER JETBLUEPRINTS accessed April 18, 2017. http : //www.newsweek.com/north− korean−hackers− steal − f15 − fighter − jet− blueprints− 470135

Esmaili, M., Balachandran, B., Safavi-Naini, R., and Pieprzyk, J. (1996, December). Case-based reasoning for intrusion detection. Computer Security Applications Conference,1996., 12th Annual (pp. 214-223). IEEE.

Johng, H. (2016, May). Estimating performance and of cloud-based systems : a model driven,complementary approach.

Lee, Y. (2016, December). N.Korean Hackers Accessed Military Intranet accessed April 18,2017. http : //english.chosun.com/site/data/htmldir/2016/12/07/2016120701429.html

Pearson, M. (2013, March). Hacking attack on South Korea traced to China, officials sayaccessed April 18, 2017. http : //www.cnn.com/2013/03/20/world/asia/south− korea−computer − outage/

Ponemon Institute (2016). 2016 Cost of Cyber Crime Study & the Risk of Business Innova-tion. Ponemon Instituite.

Simmons, C., Ellis, C., Shiva, S., Dasgupta, D., and Wu, Q. (2009). AVOIDIT: A CyberAttack Taxonomy .

Sudworth, J. (2009, July). New ’cyber attacks’ hit S Korea accessed April 18, 2017. http ://news.bbc.co.uk/2/hi/asia− pacific/8142282.stm

Supakkul, S., Hill, T., Chung, L., and Oladimeji, E. A. (2009, August). Goal-oriented securitythreat mitigation patterns: A case of credit card theft mitigation. Proceedings of the 16thConference on Pattern Languages of Programs (p. 17). ACM.

24

Page 35: Cyberattack Detection & Mitigation: A Goal-Oriented and

BIOGRAPHICAL SKETCH

Sang-woo Moon received his Bachelor of Science and Art of Military degree from Korea

Military Academy at Seoul, South Korea in 2012. He continued his Master’s degree at The

University of Texas at Dallas in 2015. His research interests include Security, Non-Functional

Requirements, and Machine Learning.

25

Page 36: Cyberattack Detection & Mitigation: A Goal-Oriented and

CURRICULUM VITAE

Sang-woo MoonApril 21, 2017

Contact Information:

Apt 1026.800 W. Renner Rd.Richardson, TX 75080, U.S.A.

Voice: (469) 900-6177Email: [email protected]

Educational History:

B.S., Computer Science and Military Arts and Science, Korea Military Academy, 2012

Employment History:

Army Officer, Republic of Korea Army, March 2012 – present