Anthropic Refuses Pentagon Demand on AI Weapons

Dario Amodei published a statement confirming Anthropic already deploys Claude across U.S. defense and intelligence agencies for planning, cyber operations, and analysis – but will not remove two guardrails the Department of War demanded it drop. The company drew hard lines against mass domestic surveillance and fully autonomous weapons, citing both civil liberties risk and insufficient model reliability. The Pentagon responded with threats to designate Anthropic a supply chain risk and invoke the Defense Production Act. It is a remarkably public standoff. OpenAI and Google have moved steadily closer to defense contracts over the past year, making Anthropic’s refusal an increasingly lonely position among frontier AI labs. Amodei offered to help transition the government to other providers if cut off, which reads less like compromise and more like calling a bluff.