Anthropic embeds six METR researchers with near-employee access and no veto on findings

Anthropic has started giving employee-like access to an embedded team from METR, a nonprofit that evaluates frontier AI models for risk. Six METR researchers now have company hardware, office space and access to internal tools and systems, roughly the same access as Anthropic's own risk teams. Staff are encouraged to share with them as they would with coworkers, with exceptions such as customer data.

The key term is publication. METR can publish what it finds without Anthropic's approval. Anthropic can only require redactions in a few narrow categories, and METR can say publicly whether those redactions were justified. Separately, the biosecurity group SecureBio is reviewing the chemical and biological weapons chapter of Anthropic's August risk report.

The move follows a September commitment by Anthropic to stronger independent assessment, and it landed the day after federal regulators sent subpoenas to Anthropic and OpenAI. Anthropic says every frontier AI company should do the same. OpenAI promised auditors similar access in September, but its relationship with METR has since been strained.

Our take

This is the strongest independent oversight any frontier lab has accepted so far, and the no-veto clause is what gives it teeth. It also sets a public bar that rivals will be measured against.