METR's first embedded eval at OpenAI: auditing agent misalignment monitoring
j_asminewang · x · 2026-10-23
Two employees from METR and Redwood Research spent three weeks on-site at OpenAI this summer in an embedded assessment pilot, evaluating the systems OpenAI uses to monitor for misalignment from internal agents; a brief write-up is now public. Former OpenAI safety leader Miles Brundage notes METR gets the best terms available while most companies have done zero embedding, arguing the field needs binding standards rather than ad hoc bargaining.
More from Safety
- Honeypot hit 14,000 times in 6 hours as autonomous AI agents swarm the internet — marcvanderchijs · 2026-10-23
- Dragonfly files CFTC comment letter urging standardized compute derivatives markets — AccBalanced · 2026-10-23
- Quest Diagnostics hit by hack, exposing patient data at a firm serving half of US hospitals — cephaloform · 2026-10-23
- OpenAI fired three researchers fighting to keep chain-of-thought visible after requesting their Signal chats — JeffLadish · 2026-10-23
- Cybersecurity is now a token-attrition war of red vs blue agent swarms — AccBalanced · 2026-10-23
- Abliterated GLM-5.3 sells for $10 as "world-class penetration tester", fueling vulnerability surge — deedydas · 2026-10-23