Anthropic gives METR embedded access for independent evaluations

Anthropic announced on October 15 that six METR staff members are receiving company hardware, office space and access to internal tools. The arrangement gives the independent evaluator a closer view of model risks and agent incidents than an ordinary external test.

METR can publish its assessments without an Anthropic veto. Anthropic retains the ability to require redactions for predefined categories of confidential information, while METR can describe limitations on the access it received.

The distinction matters: independent evaluation depends on both what investigators can inspect and what they are allowed to tell the public. Access alone would leave the second half unresolved.