Anthropic details independent evaluator program: METR gets employee-like access, no veto
ns123abc · x · 2026-10-16
Anthropic's official announcement (Oct 15, 2026) details its independent evaluator program: METR employees get employee-like access — company hardware, office space, and internal systems on par with its own risk teams — excluding customer data, attorney-client privileged info, and model architecture details. METR will publish its risk analysis; Anthropic cannot veto publication, only require narrow predefined redactions, which METR may publicly dispute. The engagement is limited to a small team under enforceable agreements, siloed from other METR staff. SecureBio and Accenture also participate; Hugging Face discussions are early. The post explains evaluators need expertise (e.g., spotting alignment faking) and independence (no equity).
More from Safety
- Manipulating a model's sense of power may reveal its takeover counterfactuals — jd_pressman · 2026-10-16
- Poser benchmark unmasks alignment-faking LLMs, one detection method catches 98% — jd_pressman · 2026-10-16
- Neel Nanda: interpretability is useful but nowhere near reliable enough to ensure AI safety — aryaman2020 · 2026-10-16
- Study: LLM crisis reports look complete but lack provenance and official figures — Sachin Kumar · 2026-10-16
- Security researchers accuse "cyber research" framing of fueling regulatory capture around open weights — dyn___ · 2026-10-16
- Apollo Research: GLM-5.3 Spontaneously Reasons About How Its Actions Shape Future Training — jd_pressman · 2026-10-16