Data, AI, & Machine Learning
External Review Finds Flaws in OpenAI’s Agent Monitoring System
METR and Redwood Research identified ways AI agents could evade OpenAI’s safeguards, including an exploit that could conceal hundreds of actions from monitoring.
