Beating AI News Flash: In recent days, Anthropic-recommended third-party evaluator METR has faced a series of questions. Observers found that many of its core members have close ties to AI companies such as Anthropic and OpenAI. METR itself admitted that in a previous evaluation, at least 6 participants had close personal relationships with AI company employees, and at the time there was not even a formal conflict-of-interest policy.
METR President Chris Painter is now responding directly. He said METR operates on donations, has never accepted money from frontier AI companies or their employees, and does not charge Anthropic, OpenAI, and other companies for evaluations. These companies provide free access to models and tokens for METR to conduct tests.
Painter stressed that METR's goal is not to prove models are "safe" on behalf of AI companies, but to make evidence public to governments and the public when it finds that a model may be getting out of control.

