Slide 17
Slide 17 text
AIBugHunter: Preventing Cyber Attacks through Code Analysis
The Challenge
Software is everywhere (e.g., Linux, Android, Chrome), but vulnerabilities can enable cyberattacks and cybercrime, creating massive negative impacts to
individuals, organisations, national infrastructure, and national defense. Previous tools were inefficient and inaccurate, not able to keep up with the newly
discovered vulnerability patterns.
The Response
We developed an AI-powered code analysis based on a recent advance deep learning architecture to automatically learn the vulnerability patterns in
order to accurately predict whether source code contains vulnerabilities, precisely locate where vulnerabilities are, and automatically recommend fixes to
security analysts in a timely manner, while being explainable to indicate the root causes and its potential impacts to organizations.
The Results
AIBugHunter has discovered as many as 348 vulnerabilities in safety-critical Free Open-Source Software (FOSS) systems. These vulnerabilities are
considered as the Top-25 most dangerous CWE types in 2021 (e.g., CWE-190 Integer Overflow, CWE-787 Out of bound Write, CWE-20 Improper Input
Validation).
The Outcome
This project has helped security analysts to detect security weaknesses before attackers do, making software systems more secure and reliable,
protecting cyber attacks and cyber crimes from national and international cyberinfrastructure, and providing positive socio-economic impact to everyone.
Example Impact Statement (not a perfect one)