Back to speakers

EX-AI-Summit · Speaker

William Berry

Independent Research | Listening Post

William Berry works on independent agent evaluation and containment research through the Listening Post project. His proposed evaluation work examines whether AI agents stay within their constraints while attempting tasks, using known-good and known-bad controls to test the benchmarks themselves.
Session

Evaluating agent containment: policy violations, tool abuse and shutdown resistance, rather than task completion alone.

We use essential cookies to run this site and optional analytics cookies to understand how it is used. You can accept all or reject optional cookies. See our privacy notice and legal.