Anthropic gives METR evaluators employee-like access — and the right to publish without its approval

Anthropic said on Thursday, Oct. 15, 2026, that it has started giving a team from METR, a nonprofit that evaluates frontier AI models, employee-like access to its systems and staff, and that METR will publish its own assessments of Anthropic's risk levels, incidents and practices.

Most "independent audits" in AI are a company deciding what an outside group gets to see and then approving the write-up. This one hands six outsiders a badge, a laptop and the internal tools — and takes away the veto. The test is the first METR report: if it lands with real findings and only the narrow redactions Anthropic promised, the bank-examiner model stops being a metaphor. Every other lab now gets asked why it hasn't done the same.

Anthropic said on Oct. 15, 2026, that it has started giving employee-like access to a team from METR, a nonprofit that evaluates frontier AI models. It cites a commitment last month to stronger independent third-party assessment, and its signing of the White House Accord on Super Intelligence, under which companies pledged to partner with independent external auditors or evaluators. Those lines are Anthropic's.

Six METR employees have been onboarded with company hardware, office space, and access to internal tools and systems roughly on par with Anthropic's own risk-assessment teams. Anthropic staff are encouraged to share information with them as they would with a fellow employee working on alignment risk, outside narrow categories such as customer data, attorney-client privileged information, and model architecture information. Alignment, here, means whether a model stays on the intended goal. Those lines are Anthropic's.

METR will publish its analysis of Anthropic's risk levels, incidents, or practices. Anthropic cannot veto publication. It can require redaction in narrow predefined categories. METR may publicly characterize those redactions and say whether it thinks they were justified, without revealing the redacted information. METR has not published findings from this arrangement yet. Those lines are Anthropic's.

The METR team works under enforceable agreements and is siloed from other METR employees, including researchers who work with Anthropic on other projects. Anthropic compares the setup to bank supervision, where regulators assign resident examiners inside the largest banks. Those lines are Anthropic's.

Anthropic describes three tiers of evaluator access. Tier one is narrowly scoped investigations, with publication only with Anthropic's approval. Tier two is more information and publication without approval, though narrow redactions are possible. Tier three is METR's arrangement: employee-like access plus tier-two publication rights. Those tiers are Anthropic's.

Other evaluators, as Anthropic lists them: SecureBio is at tier one, reviewing the chemical and biological weapons chapter of Anthropic's August Risk Report after a published pilot review. Accenture and Faculty are working toward a partnership covering evaluation, red-teaming, alignment assessment, and safeguard testing. Anthropic also says it is in early discussions with others, including Hugging Face. Red-teaming, here, means trying to break or misuse a model on purpose so weak spots show up. Those lines are Anthropic's.

Anthropic wrote: "We're embedding evaluators voluntarily, and we believe every frontier AI company should do the same." That quotation is Anthropic's.

METR's own Oct. 14 posts, as background: it has prototyped loss-of-control assessments with developers since 2025 and published a pilot Frontier Risk Report in May with Anthropic, Google, Meta, and OpenAI. Loss of control, as METR uses it, is the risk that a model could act in ways people cannot reliably stop. Those lines are METR's. They are not findings from the new Anthropic embed.

The picture is Anthropic CEO Dario Amodei, on the right, speaking onstage with an interviewer at TechCrunch Disrupt 2023. It is a Wikimedia Commons photograph. The frame does not print a calendar date.

In plain terms, Anthropic said on Thursday that six METR staff now sit inside the lab with badges, laptops, and internal tools, and that METR can publish its read of Anthropic's risks, incidents, and practices without a company veto. Anthropic can still demand narrow redactions. METR has not published that first report yet. Anthropic says every other frontier lab should do the same.