Outside evaluators find vulnerabilities in OpenAI's internal agent-monitoring system

Nonprofit evaluators METR and Redwood Research said Thursday that they spent three weeks this summer (July 21 to Aug. 11) embedded at OpenAI stress-testing its internal misalignment-monitoring system, with chaperoned access to its codebase, documentation and technical staff. They found several novel vulnerabilities, including an exploit that would let agents keep hundreds of recent actions from being monitored. OpenAI says it has patched the flaws affecting its production safeguards. The evaluators shared a 61-page internal report with OpenAI but published only a high-level summary, and the pilot follows Sam Altman's September pledge to give independent assessors employee-like access.