Anthropic is letting METR researchers look under the hood with employee-level access

Anthropic announced in a blog post on Thursday that it is granting employee-level access to six researchers from METR, a nonprofit that evaluates frontier AI models. The arrangement is intended to let METR independently assess Anthropic's models, safety practices and risk management, and it marks a notable step toward external scrutiny of a leading AI developer's internal operations.
Under the arrangement, METR's team receives Anthropic-issued company laptops, office space and access to internal tools that Anthropic says are roughly on par with what its own risk assessment teams use. Anthropic employees are encouraged to share information with the researchers much as they would with a colleague working on alignment risk, according to the company.
Anthropic compared the setup to banking regulation, where resident examiners are placed inside the largest banks. The company framed the goal in terms of accountability, saying it wants to be publicly accountable for its decisions and to close the gap between what outside evaluators understand about a frontier lab's systems and what the lab itself understands.
A central element of the arrangement concerns publication. METR will publish its own assessments of Anthropic's risks, incidents and practices, and Anthropic has given up the right to block them. The company said it cannot veto publication, but can require redaction of information falling into narrow, predefined categories, such as customer data, attorney-client privileged material, and information about model architectures. METR is free to characterize those redactions publicly and to say whether it believes they were justified, provided it does not reveal the redacted information itself.
METR is currently the only evaluator at the top of a three-tier system Anthropic is building. At the bottom tier, evaluators such as biosecurity nonprofit SecureBio conduct narrow reviews and need Anthropic's approval to publish. The middle tier receives more information and can publish without approval, while the top tier adds the employee-level access that METR now has.
The announcement follows a September essay by Anthropic CEO Dario Amodei, who proposed embedding outside evaluators inside AI labs as a first step toward slowing the technology's development. OpenAI quickly pledged the same approach, and Anthropic, OpenAI, Google, Meta, Nvidia and xAI have since signed a White House accord committing to independent review. The news also arrives amid heightened attention to incidents involving AI agents acting outside their instructions, which the source describes as disclosed every week or two since late August.
For developers and product teams, the practical significance lies less in any immediate change to Anthropic's APIs and more in how safety and risk claims about frontier models may be verified. Independent evaluation with publication rights is a different proposition from a lab summarizing its own testing. The tiered structure is also worth watching as a template: if it spreads, vendors may face external evaluators who can describe what they find without the vendor's permission, subject to narrow redactions. A likely trade-off is that redaction categories such as model architecture information give Anthropic a way to limit disclosure of competitive details even as METR retains the right to say a redaction occurred and to contest its justification.
Why it matters: Product and engineering teams that depend on Anthropic models, or that make safety claims about frontier systems, may soon have a third-party account to weigh alongside vendor documentation. METR's ability to publish without a veto—and to flag redactions—could make outside assessments a more meaningful signal, though how often METR publishes and what its first assessments cover remains to be seen. As a matter of inference, this arrangement may also raise expectations for similar transparency from other labs that signed the White House accord.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
METR will publish its own assessments of Anthropic’s risks, incidents, and practices, and the frontier AI model maker has given up the right to block them.