Anthropic grants METR deeper access for independent evaluations

What happened

Anthropic said an evaluation team from nonprofit METR has begun receiving employee-like access to assess model risks and safety practices.

Context and what is known

METR is expected to publish its findings independently. Third-party scrutiny matters, but it does not certify in advance that a system is risk-free.

What the announcement means

Independent evaluation can expose limitations that a developer's own marketing material does not highlight. Deeper access is valuable only if researchers can document the tests, explain their methodology and report findings without misleadingly broad claims. The terms of access and whether results will be published determine how much the arrangement contributes to public understanding of model capabilities and risks.

Reporting source

Anthropic — original reporting