Anthropic gives outside evaluators deeper access to check its AI risks

In short

Anthropic says it has given a six-person team from METR, a nonprofit that evaluates advanced AI models, access to its internal tools and systems that is similar to what its own staff have. The company also describes three levels of access for outside evaluators. The announcement does not report any model misbehaving.

What happened

Anthropic published the update on October 15, 2026. It says it committed last month to stronger independent, third-party assessment of its models and safety practices. It also says it signed the White House Accord on Super Intelligence, in which companies pledged to work with independent external auditors or evaluators.

Anthropic says it has now onboarded six METR employees. They have company hardware, office space, and access to internal tools and systems roughly on par with Anthropic's own risk assessment teams. A few categories are excluded, such as customer data, attorney-client privileged information, and information about model architectures.

Anthropic says METR will publish its analysis of Anthropic's risk levels, incidents or practices. Anthropic says it cannot veto publication. It can require redactions in narrow, predefined categories. METR may say publicly whether it thinks those redactions were justified, without revealing the redacted material.

Anthropic describes three tiers of access:

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

Anthropic says SecureBio is reviewing the chemical and biological weapons chapter of its August Risk Report, after publishing a pilot review of the previous report. It says it is working toward a partnership with Accenture and Faculty covering model testing, alignment assessments and safeguard testing. It is also in early talks with other groups, including Hugging Face.

What was done about it

Nothing in the report describes a problem that needed fixing. The step is a change in how Anthropic works with outsiders. Anthropic says METR's team operates under enforceable agreements, and other METR staff are kept apart from its work. Anthropic says it will keep sharing updates and hopes to fully embed more evaluators as trust builds.

Why it matters

Anthropic says it cannot publicly share everything about its safety work, because some of it could help malicious actors or is commercially sensitive. It says outside evaluators with deep access can close the gap between what they understand about a company's systems and what the company understands. It compares the idea to bank supervision, where regulators place examiners inside large banks. Anthropic says it believes every frontier AI company should embed evaluators voluntarily.

Read the original report.