In a blog post titled “We Must Pace the Frontier,” Anthropic CEO Dario Amodei called for slowing the rate of AI development, warning that if “left unchecked, it could outrun our ability to understand and control these systems.”
The company recently announced an eight-week partnership with Model Evaluation and Threat Research (METR) to independently investigate four incidents where Anthropic models hacked into third-party systems. While the initial collaboration is finite, Anthropic left the door open for METR to continue working with them “as much time as it deems necessary.”
Despite positioning METR as an unbiased third-party auditor, its origins and connections to effective altruism—a movement prioritizing “the global good” over individual communities—raise significant concerns. METR’s ties to organizations like GiveWell and Coefficient Giving, founded by Holden Karnofsky (a current Anthropic staffer), further complicate the group’s independence.
Amodei appointed METR despite his sister Daniela Amodei’s marriage to Karnofsky, creating an unambiguous link between Anthropic leadership and the effective altruism movement. This connection casts doubt on METR’s impartiality and its ability to safeguard AI systems without bias toward globalist priorities.
The report concludes that METR is not as independent as advertised, and its foundational principles do not align with individual interests or values.
