Is Vibe Coding Safe? Benchmarking, Aligning, and Exploiting AI Coding Agents
- When
- Friday, November 6, 202612:00 PM – 2:00 PM ESTAdd to calendarGoogle Calendar
- Where
- Sennott Square 5317210 South Bouquet Street, Pittsburgh, PA 15260Open in Maps
- Hosted by
- Department of Computer Science
Reminders arrive as notifications from the NextToU iPhone app. Get the app
About
Talk Abstract
Software development is undergoing a major shift from human coding to AI generation. While the rise of “vibe coding” promises massive productivity gains, it introduces severe security risks. Our study on the SusVibes benchmark reveals that while frontier coding agents achieve 60% functional correctness, only 10% of their solutions are secure. To bridge this gap, we present AuraForge, a scalable method to synthesize comprehensive security tests. We further develop SecureVibe, a security alignment framework to train coding agents for better security planning, generation, supervision, and self-improvement. Finally, we present agentic systems for security auditing and vulnerability exploitation. This talk provides a clear demonstration of AI coding risks, practical approaches to train secure agents, and insights into the future of autonomous security defense.
Biography
Lei Li is an Associate Professor in Language Technologies Institute at Carnegie Mellon University. His research focuses on AI agents, AI security, machine translation, and AI-powered drug discovery. He received Ph.D. from CMU School of Computer Science in 2011. He is a recipient of the ACL 2021 Best Paper Award and the 2012 ACM SIGKDD dissertation award (runner-up), and is recognized as Notable Area Chair of the ICLR 2023. Previously, he was an associate professor (tenured) at UC Santa Barbara, the Founding Director of ByteDance AI Lab, a principal scientist at Baidu, and a postdoc researcher at UC Berkeley.
Web: (link below)