Anthropic gives METR employee-like access to assess its AI models

GoKawiil's interpretation of the reporting above, not reported fact.

Granting deep, employee-like access suggests Anthropic is trying to narrow the gap between what outside evaluators know and what the company itself knows about its systems, which could set a precedent for how AI labs demonstrate accountability without disclosing sensitive compute or training data. The involvement of multiple evaluators, including a potential future role for Hugging Face, may signal an attempt to build a broader ecosystem of independent oversight across the industry.

• METR has been given employee-like access to evaluate Anthropic's frontier models and will publish its risk assessments.
• Anthropic is also working with SecureBio and Accenture, and is in early talks with other organizations including Hugging Face.
• The move follows Anthropic's commitment to third-party assessment and its signing of the White House Accord on Super Intelligence.