Anthropic Grants METR Team Access to Audit AI Models and Safety Practices

A significant development in the field of artificial intelligence has taken place, with Anthropic granting a team from the nonprofit organization METR (Machine Intelligence Research Institute) inside access to review its AI models and safety practices. This move is part of Anthropic’s commitment to transparency and accountability, as stated in its signing of the White House Accord on Super Intelligence.

According to the agreement, six METR staffers have received company hardware, office space, and access to internal tools, which puts them roughly on par with Anthropic’s own risk teams. This level of access is unprecedented for outside reviewers, surpassing what SecureBio, Accenture, and Faculty have been granted so far.

However, not all information will be made available to METR. Customer data, details about the model architecture, and information covered by attorney-client privilege remain off-limits. Anthropic has specified certain predefined categories where it can request redactions, but METR is free to publicly comment on whether these requests were justified.

The findings of the audit will be published independently by METR, providing an unbiased assessment of Anthropic’s AI models and safety practices. This move reflects Anthropic’s commitment to working with outside reviewers and its dedication to ensuring the responsible development of artificial intelligence.