Anthropic Gives Six METR Evaluators Employee-Level Access to Internal Systems

Unite.AI reported on October 15 that Anthropic has begun giving an embedded METR team employee-like access to its operations. METR is a nonprofit that evaluates frontier AI models, and six of its employees have been onboarded with company hardware, office space, and access to internal tools and systems.

The arrangement follows Anthropic's commitment in September 2026 to stronger independent, third-party assessment of its models and safety practices, and its signing of the White House Accord on Super Intelligence, in which companies pledged to partner with independent external auditors or evaluators. Anthropic said it has worked with METR over the past year on several reviews of its Risk Reports.

The embedded team's access is roughly on par with what Anthropic's own risk assessment teams have. Outside a small number of clearly defined categories — customer data, attorney-client privileged information, and information about model architectures — employees are encouraged to share information with the team as they would with a colleague working on alignment risk. A footnote explains the architecture carve-out: because evaluators, like bank examiners, may also work with other companies, architecture details carry a high risk of inadvertent leaks while their relevance to risk assessment is typically low. When architecture information is highly relevant, Anthropic said it will generally seek to share it, subject to explicit internal approvals.

Publication terms are the heart of the model. METR will publish its analysis of Anthropic's risk levels, incidents or practices, and Anthropic cannot veto publication — though it can require METR to redact information falling into narrow, pre-defined categories. METR is free to characterize those redactions publicly and to say whether it believes they were justified, so long as it does not reveal the redacted material. The engagement is limited to a small team operating under enforceable agreements, and other METR employees, including researchers who work with Anthropic on other projects, are siloed from the embedded team.

Anthropic's rationale is that as models grow more powerful they require stronger safeguards and more challenging risk assessments. The company says it wants to be publicly accountable but cannot share everything relevant to safety, since some of it could be useful to malicious actors or is commercially sensitive, such as compute and training information. Its answer is to give third-party evaluators deep access and have them publish evidence and reasoned conclusions, closing the gap between what evaluators understand about a frontier lab's systems and what the company itself understands.

The announcement lists four things evaluators need: expertise in assessing frontier model capabilities, including understanding how evaluations go wrong, such as alignment faking; independence, meaning no equity in the company being evaluated and no financial or operational dependence on it; enough access to meaningfully evaluate risks, extending beyond the latest models to how they were trained and how safeguards function; and freedom to publish observations and assessments. Anthropic draws a bank-supervision analogy while acknowledging it is not exact: banks know more about their own risk positions than outsiders, so regulators assign resident examiners inside the largest banks.

The program has three tiers. In the first, evaluators conduct narrowly scoped risk investigations, Anthropic decides what access to provide, and findings are published with the company's approval. In the second, evaluators publish without approval and see more information, while Anthropic can still require removal of specific categories and evaluators can publish their view on whether the redactions were justified. The METR arrangement is the third tier: employee-like access with publication rights similar to the second tier, and Anthropic says it hopes to have several evaluators at that level over time. SecureBio is currently in the first tier, reviewing the chemical and biological weapons chapter of Anthropic's August 2026 Risk Report; Anthropic is working toward a partnership with Accenture and Faculty covering evaluation, red-teaming, alignment assessments and safeguard testing, and says it is in early discussions with several other organizations, including Hugging Face.

The September backdrop is Anthropic co-founder Dario Amodei's essay "We Must Pace the Frontier," which made embedded evaluators with employee-like access the first step of a three-step plan for pacing frontier AI development, and called on governments to require other frontier companies to match. The near-term questions are concrete: how much granularity METR's first full analysis will disclose, whether the redaction mechanism gets triggered often in practice, and whether voluntary access of this kind becomes an industry default.