Abstract:
AI tools are increasingly being adopted in programming, a foundational task that shapes our digital society. However, the inherent fragility of AI brings the potential to introduce software flaws, in ways unprecedented in traditional software development. It is critical to identify and mitigate these new risks for ensuring the security and reliability of software systems. In this talk, I will first provide an overview of my research towards addressing this fundamental challenge. Then, I will delve into AI code generation, discussing its security concerns and methods to train more secure AI coding assistants. Finally, I will outline several promising directions for future research.
Speaker Bio:
Jingxuan He is a PhD candidate at ETH Zurich, advised by Martin Vechev. His research focuses on exploring the synergy of machine learning and programming, as well as its implications for security and reliability. His work has been published in top venues in PL/ML/security, recognized by a Distinguished Paper Award at CCS and an OpenAI Cybersecurity Grant, and adopted by industry (e.g., Google and Snyk).
