extractor: extracting attack behavior graphs from threat

9
EXTRACTOR: Extracting Attack Behavior Graphs from Threat Reports ksatva2, rgjome1, [email protected] University of Illinois at Chicago 1 f Kiavash Satvat Rigel Gjomemo V.N. Venkatakrishnan

Upload: others

Post on 16-Oct-2021

5 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: EXTRACTOR: Extracting Attack Behavior Graphs from Threat

EXTRACTOR: Extracting Attack Behavior Graphs from Threat

Reports

ksatva2, rgjome1, [email protected]

University of Illinois at Chicago1

f

Kiavash Satvat Rigel Gjomemo V.N. Venkatakrishnan

Page 2: EXTRACTOR: Extracting Attack Behavior Graphs from Threat

2

f

A significant amount of knowledge is available in Cyber Threat Intelligence (CTI) reports.

Cyber Threat Intelligence Reports

Page 3: EXTRACTOR: Extracting Attack Behavior Graphs from Threat

Human-intensive Effort for Creation of Detection Rules

f

Wouldn’t it be great to automate and scale?

Intrusion Detection Systems

CTI Reports

Page 4: EXTRACTOR: Extracting Attack Behavior Graphs from Threat

What is being done automatically?

• Search for fragmented Indicators of Compromise (IOC)• Hash values, file/process names, IP addresses, domain names

Limitation:• Updated or re-purposed attacks and malware polymorphism• Use of legitimate-looking names (like svchost in Windows)• Easy for the attacker to mutate and evade detection systems!

What if we can learn more than just isolated IOCs?Something which is harder to evade!...

f

Page 5: EXTRACTOR: Extracting Attack Behavior Graphs from Threat

Problem Statement

• Extract actionable provenance graphs of attack behavior from natural language CTI reports• Actionable: Provenance graphs can be directly used to perform threat

hunting• Behavior: connected events and entities, not single IOCs

f

Extractor Input

Page 6: EXTRACTOR: Extracting Attack Behavior Graphs from Threat

Challenges and Approach

f

• Verbosity

• Inter vs Intra verbosity

• Relationships Extraction (Subject, Verb, Object)

• Causality and flow of attack

• CTI Language Complexities

• Domain specific vocabulary

• Ellipsis subjects and objects

• Pronoun

• Passive vs active

EXTRACTOR

Page 7: EXTRACTOR: Extracting Attack Behavior Graphs from Threat

Evaluation

f

EXTRACTOR

8 DARPA Transparent

Computing

Engagement 2

Large-Scale Evaluation

POIROT

2.

3.

Precision Recall F1-Score

0.89 0.94 0.92

1.

Detection False Positive

0

# reports Similarity Score

Microsoft 4020 91%

TrendMicro 11600 85%

Technical description

vs

solution section

Threat Hunting Tool

EXTRACTOR

POIROT

Precision Recall F1-Score

0.96 0.93 0.94

Detection False Positive

0

Threat Hunting Tool

6 public CTI reports

EXTRACTOR

Page 8: EXTRACTOR: Extracting Attack Behavior Graphs from Threat

Sample Attack

Behavior Graphs

from Public CTI

Reports

Source code:

https://github.com/ksatvat/

EXTRACTOR

8

f

Page 9: EXTRACTOR: Extracting Attack Behavior Graphs from Threat

Questions?

f