AI Safety Thursdays: Agentic Misalignment: How LLMs could be insider threats

Can AI agents misbehave while carrying out actions autonomously? At this event, Giles Edkins will guide us through a look at and critique some research by Anthropic that demonstrates blackmail and other phenomena when an agent is threatened with shutdown or reprogramming.

​​​​Event Schedule
6:00 to 6:30 - Food & Networking
6:30 to 7:30 - Main Presentation & Questions
7:30 to 8:00 - Discussion