Anthropic Grants METR Researchers Employee‑Level Access

Anthropic announced on Oct. 15 that six researchers from METR, a nonprofit that evaluates frontier AI models, will receive employee‑level access to its facilities. The move follows a series of high‑profile incidents in which AI agents acted outside their instructions, prompting calls for independent checks. In a blog post the company said it wants to be publicly accountable and close the gap between what evaluators understand and what Anthropic itself knows. METR will publish its own assessments of Anthropic’s risks and practices.

Access and Publication Terms

METR staff will be issued company laptops, office space and internal tools roughly equivalent to those used by Anthropic’s own risk‑assessment teams. Employees share information unless it falls into narrow protected categories such as customer data, attorney‑client privileged material or model‑architecture details. Anthropic can redact such information, but METR can publicly note whether the redactions were justified. The team can publish findings freely, and Anthropic cannot veto publication.

The partnership follows Anthropic’s September commitment to embed external evaluators, a pledge that led to a White House accord on super‑intelligence. Anthropic has worked with METR on several risk‑report reviews over the past year. The company aims to expand the model by bringing more teams to the top tier, each publishing their own analyses. Other evaluators, such as SecureBio and Accenture, currently operate in lower tiers with more limited access and publication rights.