The why and how of AI Compute Verification - Yannick Mühlhäuser
Summary
While verification of AI agreements is still emerging, it is already established practice in other high-stakes scenarios. Examples include: Nuclear inspectors monitor fissile material, and chemical weapons treaties come with on-site inspections.
For AI, comparable verification tools are only now being developed.
Yannick will talk about:
why states and AI companies are interested in technical options for verifying AI agreements,
what verification could look like in practice, and what lessons carry over from nuclear and chemical governance, and
where the field stands today and what to expect next.
Format
📅 Wednesday, 4th Nov, 18:00 - 19:00 CEST
Online talk, followed by Q&A via Slido
Speaker Bio
Yannick Mühlhäuser is co-founder of the Verifiable Compute Foundation, a research affiliate at the Oxford Martin Hardware AI Governance Initiative and a policy researcher at the Future of Life Institute. He is also a co-founder of the Frontier AI Security Residency. His work focuses on AI compute verification for AI agreements, including cryptographic techniques and methods for detecting hidden compute infrastructure. He holds bachelor’s and master’s degrees in physics from the University of Tübingen, where his graduate work focused on the interpretability of neural networks.
Who should attend?
Anyone interested in technical AI governance, especially students and researchers in relevant areas such as policy, international relations, law or computer science. No technical background needed.