Anthropic grants employee like access to METR for frontier AI model evaluation; partners with SecureBio and Accenture (ACN); is in talks with Hugging Face for evaluation

Frontier labs opening their evaluation stack to third-party bodies like METR fits a pattern that has run through the sector: external model assessment has become the de facto substitute for formal AI regulation, and labs have historically courted evaluators early to shape the methodology they will eventually be judged against. The Anthropic posture here is consistent with its long-standing positioning as the safety-forward lab, a brand that has been commercially useful in winning enterprise and government-adjacent business. The tie-ups split into two channels worth distinguishing: SecureBio and METR speak to the capability-evaluation and bio-risk side, which feeds the safety-policy narrative in Washington, while Accenture is a distribution and services relationship, the channel through which foundation models actually reach enterprise budgets. The Hugging Face talks, if concluded, would matter more for open-benchmark credibility than for revenue. For ACN specifically, partnerships of this kind have tended to be directional sentiment items rather than earnings events; the precedent is that consultancies monetise AI through implementation services well after the headline. What follows worth watching is whether any evaluation results get published, since that is the point at which access arrangements become substantive.