Anthropic CEO proposes third-party AI safety evaluators linked to Effective Altruism network

Anthropic chief executive Dario Amodei said the company will allow an external nonprofit, Model Evaluation and Threat Research (METR), to act as an "embedded third-party evaluator" with employee-level access to its AI systems. In an essay titled "We Must Pace the Frontier," Amodei argued that such oversight could help prevent advanced models from causing large-scale harm.

METR, which grew out of the Alignment Research Center (ARC), has connections to the Effective Altruism (EA) community. Several of its staff, including founder Beth Barnes and chief scientist Hjalmar Wijk, previously worked with ARC, and its board includes individuals linked to EA donors such as Dustin Moskovitz and Jaan Tallinn. A METR spokesperson said the organization discloses conflicts of interest and does not accept funding from AI companies or their employees.

Critics have raised concerns about the independence of METR, noting its financial ties to donors who also back Anthropic. Perry Metzger, chairman of the Alliance for the Future, called the arrangement "a complete joke" and suggested it could create a conflict of interest. Former White House AI adviser David Sacks also questioned METR’s independence, saying the group appears to be closely aligned with Anthropic’s investors.

The proposal comes amid broader industry debate over AI safety. OpenAI CEO Sam Altman expressed support for independent evaluators with employee-like access, while some policymakers, including FTC Chair Andrew Ferguson, warned that granting AI firms coordinated safety authority could raise antitrust concerns. Anthropic declined further comment on the plan.