18
AI Red-Teaming of Software Engineering Agents
- Topic: Emerging Technologies
- Time: 18.02.2025, 14:00 to 14:30
- Location: Tech Stage
- Language:
AI-based software engineering agents are in the process of generating the majority of future source code and fixing many existing vulnerabilities. It is therefore crucial to secure these agents, focussing on protecting their AI/LLM components from adversarial threats. This talk will present findings from AI red-teaming of open source software development agents, including the main components of modern software engineering agents, risks from adversarial attacks on AI/LLM components, and promising defences.
Speaker
Dr. Beat Buesser
Beat Buesser is a researcher in the field of AI security and privacy at IBM Research in Zurich. His research focuses on evaluating the robustness against adversarial threats to machine learning and AI. He is the maintainer and leading developer of the Adversarial Robustness Toolbox (ART) and a voting member of the Linux Foundation AI & Data’s Technical Advisory Committee.