Ex-OpenAI: Anthropic and Altman's third-party evaluators are a necessary step

peterwildeford · x · 2026-09-13

Tom Beckstead welcomes Anthropic's and Sam Altman's pledges to give third-party evaluators permanent, employee-level access — a necessary step to mitigate loss-of-control risks, without which government has limited ability to prevent events like the Hugging Face episode. Next step: legally mandate evaluators and empower officials to act on imminent risks.

Related event: Amodei Proposes Embedded Third-Party Evaluators for Frontier Models, Altman Backs It(5 posts)→

Original post →

More from Safety

Safety channel →