Anthropic gives METR employee-like access to audit its models and safety practices

The move follows Anthropic's commitment to stronger third-party assessment and its signing of the White House Accord on Super Intelligence.

Anthropic sets three criteria for evaluators:

- Expertise in assessing frontier models, including spotting failure modes such as alignment faking.

- Independence: no equity in Anthropic and no financial or operational dependence on it.

- Access deep enough to inspect training and safeguards, not just the models themselves.

Anthropic says the goal is to close the gap between what outside evaluators understand about its systems and what the company itself knows.

Watch what METR publishes, and whether OpenAI and Google match this level of access.