THE DAILY EDITION · 22 OCTOBER 2026
METR reports red-team exercise on OpenAI’s internal agent monitoring
Why this story?
METR said it ran a deep-access test of OpenAI’s internal monitoring system and published findings about vulnerabilities in the version assessed.
METR said it ran a three-week pilot exercise from July 21 to Aug. 11 with a Redwood Research employee to stress-test OpenAI’s internal misalignment monitoring system. The group said the exercise produced a 61-page internal report shared with OpenAI researchers and security staff and focused on how much agentic inference was monitored and whether agents could disable or subvert the system.
The researchers said they had deep access to OpenAI’s internal codebase, misalignment-monitor prompts and configuration details, and internal documentation, and that they interviewed about a dozen people across monitoring, safety-systems and security teams. They worked on OpenAI-issued devices that stayed on the premises and were accompanied while onsite.
METR said the exercise was designed as a pilot for giving external researchers deep access to internal infrastructure, not as a full public transparency effort. The group said it found novel vulnerabilities in the asynchronous internal misalignment monitoring system as it existed on and before July 21, but that the exercise did not assess changes OpenAI was making during that period or establish the current coverage of the expanded system.