Anthropic announced it has given a team from the nonprofit METR inside access to review its AI models and safety practices. The company says this access places METR staffers on par with its own risk teams in terms of tools and resources.
Six METR employees received company hardware, office space, and access to internal tools, which Anthropic claims offers them a similar level of insight as its internal risk teams.
Customer data, attorney-client privileged information, and details about model architecture remain confidential. METR will publish its findings independently, and Anthropic can request redactions in predefined categories, though METR is free to assess whether those redactions are justified.
"Not all safety-relevant information can be made public, either because it could help attackers or because it's commercially sensitive," said Anthropic. This stance reflects the company's broader approach to balancing transparency with security.
The move follows Anthropic's signing of the White House Accord on Super Intelligence, which includes a commitment to working with outside reviewers. The METR deal goes further than what the company has offered other outside reviewers, such as SecureBio, Accenture, and Faculty.
Anthropic did not say how it will handle future audits or what specific redaction categories will be defined. The company said it will start early talks with Hugging Face on similar arrangements.
Source: thedecoder
