OpenAI hosts METR and Redwood for three-week embedded eval of internal monitoring
ShakeelHashim · x · 2026-10-23
OpenAI announced a pilot embedded evaluation this summer, inviting METR and Redwood Research on-site for three weeks with deep access to its infrastructure. Two evaluators assessed the internal system OpenAI uses to monitor for misalignment from internal agents.
Findings helped strengthen safeguards as OpenAI expanded the system and will shape future collaborations with METR and other third-party evaluators. METR has published a brief write-up. It is the first embedded third-party assessment of a frontier lab's internal alignment monitoring system.
Related event: OpenAI lets METR embed for three weeks to assess internal agent monitoring(3 posts)→
More from Safety
- OpenAI redacted 23 of 61 pages in METR's embedded misalignment-monitor report, researchers say — peterwildeford · 2026-10-23
- After eMudhra fiasco, AI agents set to audit certificates from 23 Indian CAs — kingslyj · 2026-10-23
- Security legend Halvar Flake: Rejecting AI-assisted bug reports over their origin is out of place — mboehme_ · 2026-10-23
- 8B safety classifier trained on 1,275 prompts beats GPT-5.4 across 5 benchmarks — HaggaiMaron · 2026-10-23
- DeepMind's Geoffrey Irving: alignment may need cryptography-grade rigor as AI finds 12 OpenSSL zero-days — geoffreyirving · 2026-10-23
- DeepMind's Irving calls to halt AI progress now, with treaties to follow — davidmanheim · 2026-10-23