Anthropic gives METR employee-like access after the nonprofit's embedded-assessor proposal
What's going on? On 14 October 2026, METR, the nonprofit that evaluates frontier AI systems, published a concrete proposal for "embedded assessments": independent researchers would work inside frontier labs with access to internal data, interview staff privately, and write public reports on whether developers might lose control of their AI systems.
Follow-through (15 Oct): Anthropic announced it has started giving employee-like access to a six-person METR team — company hardware, office space and internal tools roughly on par with Anthropic's own risk-assessment staff. Anthropic says it cannot veto METR's publication of assessments of risk levels, incidents or practices, but can require narrow redactions (for example customer data, attorney-client material, or model-architecture details); METR may publicly characterise whether those redactions were justified. METR sits alone for now at the top of Anthropic's three-tier evaluator ladder; SecureBio, Accenture and Faculty are named at lower or early-stage tiers, with Hugging Face among organisations in early discussions. Anthropic says it is not funding METR's embedded work.
The three questions METR wants answered: (1) alignment — what misaligned actions agents might pursue, and whether training suppresses signs of worse behaviour; (2) control — whether monitoring would catch harmful actions before they cause damage; and (3) AI R&D acceleration — whether automation of AI research itself is speeding capabilities faster than labs can understand and safeguard them.
What METR asked labs for (14 Oct proposal): Time-limited access to relevant systems and datasets on company devices, private staff interviews without a company minder present, and editorial independence to publish within an agreed scope — subject to redactions that METR says should be described transparently.
Why it matters: Several frontier CEOs floated embedding outside evaluators, and labs signed the White House Accord on Super Intelligence committing to independent assessments. METR's 14 October playbook plus Anthropic's 15 October onboarding turn that political signal into a live, bank-examiner-style arrangement inside at least one frontier lab — with publication rights that are stronger than a company-vetted safety blog post, but still stop short of true employee-level access for the whole of METR or a regulatory go/no-go gate.
Caveats: The METR proposal remains voluntary elsewhere. Anthropic's arrangement is limited to a siloed METR team under enforceable agreements; other METR staff are walled off. Architecture details may still be withheld. METR itself stresses the science is immature and these exercises cannot provide finance-style assurance against loss of control.
Sources: METR proposal (14 Oct 2026); companion posts on control and AI R&D acceleration; Anthropic update on independent evaluators (15 Oct 2026); TechCrunch; CNBC.
