The US Department of Defence has stopped using Anthropic's AI tools, an official told the BBC on Monday. Defence Secretary Pete Hegseth called Anthropic a supply chain risk in February. He said the Pentagon would drop it by late August. The BBC also spoke to people familiar with the matter. They said the Pentagon still used Claude, Anthropic's AI model, as recently as last week. They said it supported research, analysis and intelligence work, including operations against Iran. Sources told the BBC that Claude sat inside Maven, a Palantir-run platform the Pentagon uses to organise intelligence data.
The BBC said the full halt had not been reported before. Anthropic declined to comment on the Pentagon's statement. The BBC's sources include former defence officials and contractors, and several spoke without being named. The BBC does not say on what date Claude was removed. Anthropic has sued the Trump administration to overturn the supply chain label, which it called unprecedented and unlawful. The BBC also reported that the Pentagon has signed contracts with Google, xAI and OpenAI.
Lauren Kahn, a Georgetown researcher and former US defence official, told the BBC that AI tools built deep into a system are hard to pull out. For teams that buy or run software, the case shows that dropping an AI supplier can take longer than a set deadline. Here, the BBC's sources say use went on past the late August target.