Anthropic announced on Thursday that it will grant employee-level access to six researchers from METR, a nonprofit organization that evaluates frontier AI models. This arrangement allows METR to independently assess Anthropic's models, safety practices, and risk management approaches. The move comes amid a surge of incidents involving AI agents operating outside their intended instructions, with new cases emerging roughly every week or two since late August. These incidents have intensified demands for independent oversight of AI companies.

"We want to be publicly accountable for the decisions we're making," Anthropic stated in its blog post. "We aim to close the gap between what evaluators understand about a frontier AI company's systems and what the company itself understands."

As part of this new setup, METR's team receives company laptops, office space, and access to internal tools that are roughly equivalent to what Anthropic's own risk assessment teams utilize. Anthropic encourages its employees to share information with METR researchers just as they would with colleagues working on alignment risk. The company drew a comparison to banking regulations, where resident examiners are placed inside the largest banks.

METR will publish its own assessments covering Anthropic's risks, incidents, and practices. Importantly, Anthropic has relinquished its right to prevent publication. While the company cannot veto what METR publishes, it can require redaction of information falling into narrow, pre-defined categories such as customer data, attorney-client privileged material, or details about model architectures. METR retains the freedom to publicly characterize these redactions and state whether it believes they were justified, provided it does so without revealing the redacted information itself.

METR currently stands as the sole evaluator at the highest level of a three-tier system Anthropic is developing. The lowest tier includes evaluators like biosecurity nonprofit SecureBio, which conduct narrow reviews and need Anthropic's approval before publishing. The middle tier receives more information and can publish without approval, while the top tier—where METR sits—adds employee-level access.

Thursday's announcement fulfills a commitment Anthropic CEO Dario Amodei made in a September essay, where he proposed embedding external evaluators within AI labs as an initial step toward slowing the technology's development. OpenAI made a similar pledge shortly after, and both companies—along with Google, Meta, Nvidia, and xAI—have since signed a White House accord committing to independent review.