METR Faces Scrutiny Over Ties to Anthropic and Other AI Firms

Anthropic CEO Dario Amodei has argued that third-party evaluators such as METR should be embedded long term inside frontier AI companies with near employee-level access to review safety measures, incidents, and model training. According to ChainCatcher, a New York Post investigation found METR’s ties to the AI safety community are much closer than those of a typical outside auditor. METR was originally known as ARC Evals and was incubated at the Alignment Research Center founded by Paul Christiano. Christiano previously worked with Amodei at OpenAI and later became one of the first five trustees of the Anthropic Long-Term Benefit Trust. METR technical staffer Ajeya Cotra is Christiano’s wife and previously oversaw AI safety grants at Coefficient Giving. When METR evaluated Anthropic, OpenAI, Google, and Meta this year, it did not yet have a personnel conflict-of-interest policy in place or a formal recusal process. METR disclosed that at least six employees and collaborators directly involved in its projects had close personal relationships with employees at AI companies. METR said it does not accept funding from frontier AI companies or their employees and does not charge for risk assessments, though those companies provide large amounts of model tokens for free.