Anthropic gives METR evaluators employee-like access and the right to publish

Anthropic says it has onboarded six staff from METR, a non-profit that evaluates frontier AI models, with company hardware, office space and access to internal tools roughly on par with its own risk assessment teams. Customer data, privileged legal information and model architecture details are excluded.

METR will publish its analysis of Anthropic’s risk levels, incidents and practices. Anthropic says it cannot veto publication but can require redaction of predefined categories, such as confidential intellectual property.

The company describes three tiers of evaluator access: narrowly scoped reviews published with its approval; broader access with publication without approval; and the embedded, employee-like arrangement now in place with METR. It also names SecureBio, reviewing the chemical and biological weapons chapter of its August risk report, and a partnership with Accenture and Faculty.

Editor’s note: this summary is based on the company’s own announcement.