Description AI Safety attention has increasingly shifted toward “defense in depth”: a layered strategy of safeguards.
In this context, today’s event will have Quinn Dougherty leading us on an exploration of forgotten literature on proof-carrying code to leverage it in a prototype AI containment procedure, enabling a narrow form of enforceable safety.
We’ll look at preliminary experimental results, discuss strategic outlook, and cover when proof-carrying code and guaranteed-safe AI more broadly make sense. Event Schedule
6:00 to 6:45 - Pizza & Networking 6:45 to 8:15 - Main Presentation & Questions 8:15 to 9:00 - Breakout Discussions
AI Safety Thursdays—Formal Containment: Proof-Carrying Code and AI Safety
Description
AI Safety attention has increasingly shifted toward “defense in depth”: a layered strategy of safeguards.
In this context, today’s event will have Quinn Dougherty leading us on an exploration of forgotten literature on proof-carrying code to leverage it in a prototype AI containment procedure, enabling a narrow form of enforceable safety.
We’ll look at preliminary experimental results, discuss strategic outlook, and cover when proof-carrying code and guaranteed-safe AI more broadly make sense.
Event Schedule
6:00 to 6:45 - Pizza & Networking
6:45 to 8:15 - Main Presentation & Questions
8:15 to 9:00 - Breakout Discussions