Judge says Pentagon lacks Anthropic proof
A federal judge said the Pentagon lacks evidence Anthropic could sabotage delivered AI models, calling its 'supply-chain risk' rationale troubling.
U.S. District Judge Rita Lin said at a Thursday hearing that the Trump administration still has not produced evidence to justify labeling Anthropic a “supply-chain risk,” the designation the Department of Defense used to try to ban federal use of Claude. Lin, who issued a temporary block on the ban in March, is now deciding whether to make it permanent.
The Pentagon’s argument was that Anthropic’s public criticism of the DOD, over its refusal to let Claude be used for mass surveillance of Americans or for targeting and lethal-weapons decisions, justified treating the company as a security risk to the government’s own AI supply chain. Judge Lin pushed back directly, saying she saw no evidence Anthropic could alter a model already delivered to a customer or “flip some kind of kill switch,” and calling the government’s theory “really troubling” because it could let an administration punish any contractor that disagrees with it. Anthropic has maintained the underlying policy position: it isn’t the AI models that aren’t ready for those military use cases, but the surrounding safeguards and doctrine.
A separate related suit is still proceeding in Washington. If Lin’s block becomes permanent, it sets a concrete marker for how far a government customer can go in penalizing a vendor over its stated product-safety limits rather than any demonstrated technical failure.
For any operator selling into government or other security-conscious enterprise buyers, this is worth reading alongside the cybersecurity evaluation incidents Anthropic disclosed the same week, and the earlier reporting that the Pentagon demanded Anthropic enable weapons use cases it had refused: vendor risk labels are being litigated in both directions right now, and neither a real security incident nor a political dispute over acceptable-use policy is automatically dispositive on its own. The guide on government pressure on AI suppliers covers how to weigh that exposure if you sell into regulated or public-sector buyers.