Dissertation Title: Private, Verifiable, and Auditable AI Systems

Abstract: 

Designing AI systems with a balance of privacy, verifiability, and auditability is a critical foundation for a safe and trustworthy future. This research is not about model alignment but rather the ugly details of infrastructure that help make AI pipelines secure. We will draw on advancements in cryptography and security to identify risks across the AI ecosystem, from pre-training and data collection to inference and RAG, and finally to agent infrastructure. You will learn about AI pipelines, cryptography, legal risks, and regulatory frameworks for security and privacy in AI.

Committee members: 

Sandy Pentland, Professor at MIT Media Lab
Michiel Bakker, Professor at MIT Sloan & IDSS and Senior Research Scientist at Google DeepMind
Glen Weyl, Research Lead at Microsoft Research Special Projects