Scalable Formally-Verified Code Generation: Self-Improving Defenses Against Vulnerability Propagation in AI

The future of science increasingly runs on AI-generated code executed with minimal human oversight, risking the spread of vulnerabilities at unprecedented scale. Souti Chattopadhyay from the University of Southern California is building a system that compiles protective constraints into formal automata and enforces them as code is generated, rendering entire classes of vulnerabilities mathematically impossible for AI to produce.

Biography

What if entire classes of security bugs were mathematically impossible for AI to write? Souti “Rini” Chattopadhyay is a WiSE Gabilan Assistant Professor of Computer Science at USC’s Viterbi School of Engineering and founder of the Adaptive Computing Experiences (ACE) Lab. A PhD from Oregon State working at the intersection of software engineering, HCI, and cognitive science, she has earned multiple ACM/IEEE distinguished-paper honors — and is now compiling protective constraints into formal automata enforced as AI generates code.