A federal judge has blocked the Trump administration’s attempt to classify AI company Anthropic as a national-security “supply-chain risk”, finding that the move was unconstitutional retaliation following a dispute over military use of its models. The ruling removes the resulting federal-contracting restrictions and sanctions, although it does not require the Pentagon to use Anthropic’s technology or prevent it choosing rival systems.

US District Judge Rita Lin vacated Defence Secretary Pete Hegseth’s 27 February designation and an order barring military contractors and suppliers from doing business with Anthropic, describing the measures as unlawful and baseless. The dispute concerned a $200 million Pentagon deal for Anthropic’s Claude models: Anthropic sought limits on deployment in lethal autonomous weapons and mass-surveillance systems, while the Pentagon demanded permission for all lawful uses; the department is expected to appeal.