Anthropic opens models to external safety auditors

The detail worth sitting with is that METR gets direct model access, not just post-hoc audit rights. That distinction matters because it shifts the evaluator from a credentialing body into something closer to a continuous monitor, which is a meaningfully different accountability structure than anything a lab has voluntarily accepted before.

Modelwire has no prior coverage to anchor this to directly, so the honest framing is that this story belongs to a slow-building thread across the broader industry: the gradual collapse of the argument that self-regulation is sufficient for frontier labs. Anthropic is essentially conceding that point publicly and trying to own the transition rather than resist it. The competitive implication is real: if third-party access becomes a de facto industry norm, labs with less mature internal safety infrastructure face a harder path to credibility with regulators and enterprise customers alike.

Watch whether OpenAI or Google DeepMind grant METR or a comparable evaluator equivalent direct-access rights within the next six months. If neither does, Anthropic's move reads as genuine differentiation; if one follows within that window, it signals the standard is becoming table stakes rather than a strategic edge.