Anthropic embeds six METR evaluators with employee-level access, can't veto their findings
ns123abc · x · 2026-10-16
Anthropic has onboarded six METR employees with company hardware, badges, and internal systems access roughly on par with its own risk assessment teams (excluding customer data, privileged info, and model architecture details). Crucially, Anthropic cannot veto METR's publications on Claude's misalignment and risks — it can only require narrow, predefined redactions, and METR may publicly dispute whether those redactions were justified. The move implements pledges from the White House Accord on Super Intelligence; SecureBio and Accenture are also evaluators, with Hugging Face in early discussions.
More from AGI Musings
- a16z's call for startups: computer use agents that actually finish the job — andrewchen · 2026-10-16
- Engineers climb the abstraction ladder: asking for $10K of Anthropic tokens, not headcount — curious_vii · 2026-10-16
- Worldview maps the AI debate: interview pins your position against researchers and lab leaders — maxsloef · 2026-10-16
- "Cyborgs suddenly dominate math research": AI-assisted proofs spread fast — IgorCarron · 2026-10-16
- False Collatz Proof via a Lean Bug: Talk Probes Proof Assistants as LLM-Found Bug Targets — marc_lelarge · 2026-10-16
- Pennsylvania residents reject $10,000-per-household offer for massive Amazon data center — Polymarket · 2026-10-16