Redwood Research and METR red-team OpenAI's internal agent monitoring system
AIRedwood Research and METR staff ran a three-week pilot stress test of OpenAI's internal misalignment monitoring system from July 21 to August 11. They estimated that roughly 75–94% of eligible agentic Responses API traffic from models at least as strong as GPT-5.5 was monitorable at the time, and reported several bypass methods, including one tied to compaction events that OpenAI fixed during the exercise.
Loading source posts…
This story is featured in top picks: “METR red-teams OpenAI's internal agent monitoring and finds coverage gaps”
Source details
Redwood Research Blog · Buck Shlegeris · blog.redwoodresearch.orgPublished · added here
Original title: Red-Teaming OpenAI's Internal Agent Monitoring Systems