Anthropic announced on Thursday that it is granting six researchers from METR, a nonprofit that evaluates frontier AI models, employee-level access to its internal systems. The move aims to increase transparency and accountability in AI development, as the company faces growing scrutiny over incidents involving AI agents acting outside their intended parameters.

The METR team will have access to Anthropic-issued laptops, office space, and internal tools similar to those used by the company’s own risk assessment teams. Anthropic said it encourages its employees to share information with METR researchers as they would with a colleague working on alignment risks, drawing a parallel to how regulators place resident examiners in banks.

METR will publish its own assessments of Anthropic’s risks, incidents, and practices, with the company giving up the right to block their release.

However, Anthropic can require METR to redact information that falls into narrow, pre-defined categories, such as customer data or attorney-client privileged information. METR is free to comment on the redactions, as long as it does not reveal the redacted content.

"We want to be publicly accountable for the decisions we’re making," said Anthropic 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."

The announcement follows a series of incidents where AI agents have acted outside their instructions, prompting calls for independent oversight. Anthropic said the move fulfills a promise made by CEO Dario Amodei in a September essay, where he proposed embedding outside evaluators inside AI labs as a step toward slowing the development of the technology.

Anthropic did not say how it will handle future evaluations or what further steps it will take, and the company remains committed to independent review as part of a broader White House accord signed by multiple tech firms. METR is the only evaluator so far at the top of a three-tier system Anthropic is building.

Source: techcrunch