Invited Talk
in
Workshop: Red Teaming GenAI: What Can We Learn from Adversaries?

Invited Talk 7: Max Kaufmann on Red-teaming AI systems in government

Max Kaufmann

2024 Invited Talk
in
Workshop: Red Teaming GenAI: What Can We Learn from Adversaries?

Abstract

The UK government's AI Safety Institute was set up a year ago to push forward the technical capacity of the public sector. Since then they've hired ~60 technical staff and ran over 16 safety evaluation on frontier AI systems, including pre-release tests of o1 and Claude 3.5 sonnet. This talk lays out many of the technical and organisational learnings from doing that work, as well as a range of open technical questions to help push it forward.

Video

Chat is not available.