Access for outside evaluation
Anthropic said it is giving a team from METR, a nonprofit that evaluates frontier AI models, employee-like access so it can assess the safety of its systems.
The company is also in early discussions with several other organizations, including Hugging Face, about similar access. The goal is to keep safety evaluation from staying entirely inside the company.
Why it matters
As frontier models grow more capable, whether safety claims can be verified by independent organizations has become central to the debate. Employee-level access lets outside evaluators examine a model under conditions close to real use.
The update appears on Anthropic's policy page, alongside recent posts on its usage policy, its stance on open-weights models and AI regulation.
