Anthropic's designated 'independent third party' faces scrutiny: founding circle's ties with the company are deeply intertwined.

Beating AI News Flash: Anthropic CEO Dario Amodei recently advocated for third-party evaluators like METR to be embedded long-term in frontier AI companies, granted near-employee-level access to inspect safety measures, incidents, and model training processes. But a New York Post investigation found that METR and Anthropic's AI safety circle are far closer than a typical "external audit."

METR was originally called ARC Evals, incubated at the Alignment Research Center founded by Paul Christiano. Christiano not only once worked with Amodei at OpenAI—the New York Post says the two were also roommates; he later became one of the first five trustees of Anthropic's Long-Term Benefit Trust. METR's current technical staffer Ajeya Cotra is Christiano's wife and previously oversaw AI safety funding at Coefficient Giving.

A more direct problem comes from METR itself. When it evaluated Anthropic, OpenAI, Google, and Meta this year, it had no applicable personnel conflict-of-interest policy at the time and no formal recusal process. METR disclosed that among employees and collaborators directly involved in the project, at least 6 had close personal relationships with AI company employees.

Of course, this does not prove that METR is controlled by Anthropic. METR explicitly states that it does not accept funding from frontier AI companies or their employees and does not charge for risk assessments, though these companies do provide large amounts of model tokens for free.