Update
OpenAI Patches Monitoring Gaps
OpenAI has patched monitoring gaps found by outside researchers from METR and Redwood Research. The review, which lasted from July 21 to August 11, 2026, identified vulnerabilities in OpenAI’s misalignment monitoring system. The system is used to catch AI agents behaving in unintended ways. The vulnerabilities have been addressed, and the findings were shared with OpenAI researchers and security staff. The main goal was to estimate the fraction of agentic inference that gets monitored and to determine if agents could disable or subvert the monitoring entirely.