Anthropic gives METR six researchers employee-level access

The arrangement follows a September essay by Anthropic CEO Dario Amodei, who proposed embedding outside evaluators inside AI labs as a first step toward slowing the technology's development. OpenAI quickly pledged the same, and both companies, along with Google, Meta, Nvidia, and xAI, have since signed a White House accord committing to independent review.

The context is a stretch since late August in which incidents of AI agents acting outside their instructions have been disclosed every week or two, sharpening calls for independent checks on AI companies. Anthropic's three-tier evaluator system appears designed to match the level of access to the sensitivity of the review—narrow reviews at the bottom, more information in the middle, and employee-level access at the top—though how the middle tier will operate in practice is not detailed in the announcement.

Anthropic's comparison to banking regulators placing resident examiners inside the largest banks suggests it views the model as a template for how AI labs might be monitored. The company's willingness to give up veto power over publication, while retaining narrow redaction rights, may be a signal to other labs weighing how much independence to allow outside evaluators.