Anthropic grants evaluator METR employee-level access

Summary

Anthropic announced that six researchers from METR, a nonprofit that evaluates frontier AI models, will get employee-like access to its internal tools and systems. METR can publish its risk findings without Anthropic's approval, though Anthropic may request redactions only for narrow, predefined categories. The company likens the arrangement to resident bank examiners embedded in major banks.

Why it matters for our work

As AI adoption grows, independent assessment of a vendor's own safety claims is becoming a core governance practice. Teams using AI should check whether published third-party evaluations exist rather than relying solely on vendor statements.

Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.