Anthropic CEO pitches EA-linked AI watchdog to pace frontier safety

Anthropic CEO Dario Amodei proposes creating Model Evaluation and Threat Research (METR), a third-party watchdog that would have employee-like access to oversee safety practices, model training, and incident reporting across AI labs. METR traces its roots to the Effective Altruism movement and is funded by EA-aligned donors, raising concerns about independence and conflicts of interest from critics who question whether a group tied to AE figures and investors can be truly impartial. Amodei argues the measure would help regulate progress and improve transparency, while regulators and industry peers weigh in with mixed reactions as debates over AI safety governance continue.