Anthropic gives METR employee-like access, sets three evaluator tiers

Anthropic said on 15 October 2026 that it has started giving a team from METR, a nonprofit that evaluates frontier AI models, "employee-like access" to its systems. In a post titled "An update on independent evaluators", the company says METR will publish the results of its risk assessments. Everything below is Anthropic's own description; METR's account had not been published when I read the post.

What "employee-like access" means

Anthropic says it has onboarded six METR employees with company hardware, office space, and access to internal tools and systems roughly on par with what its own risk assessment teams have. Outside a small number of defined categories, it says, its staff are encouraged to share information with them as they would with a colleague. The examples it gives are data about customers, attorney-client-privileged information and information about its model architectures.

Anthropic says it cannot veto publication but can require METR to redact information in those narrow categories. METR may characterise the redactions publicly, and say whether it thinks they were justified, without revealing what was removed. The company also says the arrangement covers a small team under enforceable agreements, with other METR staff siloed from its work.

Three tiers of access

• Tier one: evaluators run narrowly scoped risk investigations. Anthropic decides what access to give, and findings are published with its approval.
• Tier two: evaluators can publish without Anthropic's approval and get access to more information. Anthropic can still require removal of specific categories.
• Tier three: the METR arrangement, with employee-like access and publication rights similar to tier two.

According to the post, SecureBio is in tier one. It is reviewing the chemical and biological weapons chapter of Anthropic's August Risk Report, after publishing a pilot review of the previous report. Anthropic says it is also working with Accenture and Faculty toward a partnership that includes evaluating and red-teaming models, conducting alignment assessments and testing safeguards, and is in early talks with other organisations, including Hugging Face.

The independence tests, and what is not said

Anthropic lists what it says evaluators need: expertise, independence (no equity in, and no financial or operational dependence on, the company evaluated), enough access, and freedom to publish. It compares the model to bank supervision, where regulators place resident examiners inside the largest banks, while noting the analogy is not exact.

The post does not say what METR is paid or by whom, when its first assessment of Anthropic will appear, or what Anthropic will do if an assessment is unfavourable. Accenture, a commercial partner, is listed as an evaluator, and the post's own criteria say evaluators should not be financially dependent on the company; how that applies to a partnership like this is not explained. Anthropic also says it signed the White House Accord on Super Intelligence, in which companies pledged to partner with independent external auditors or evaluators.

What a buyer can take from it

For a team writing vendor-risk questionnaires, the post offers three questions that apply to any model provider: what tier of access do outside evaluators have, who decides what they can publish, and what categories can the vendor redact. Until METR's reports exist, treat this as a stated commitment, not evidence about how safe any model is.