Anthropic is letting METR researchers look under the hood with employee-level access
techcrunch ·
Anthropic has granted employee-level access to six researchers from the nonprofit METR, allowing them to independently assess the company's AI models and safety practices. This move aims to increase public accountability and bridge the gap between external evaluators' understanding and internal knowledge of frontier AI systems. The arrangement provides METR with company laptops, office space, and access to internal tools similar to those used by Anthropic’s own risk teams. METR will publish its own assessments without Anthropic having veto power, though the company can require redactions for specific sensitive categories like customer data or model architecture details. This initiative fulfills a promise made by CEO Dario Amodei to embed outside evaluators within AI labs, a step intended to help slow technological development. It also aligns with recent White House accords signed by major tech firms committing to independent review. The decision comes amid growing concerns over AI agents acting outside their instructions and calls for stricter oversight.
techcrunch