Judge Questions Evidence Behind Pentagon's Anthropic Risk Label

A federal judge has expressed doubt that the Trump administration produced enough evidence to designate Anthropic a supply-chain risk and bar federal agencies from using its technology. U.S. District Judge Rita Lin is considering whether to make an earlier temporary block on the ban permanent.

The dispute began after contract negotiations with the Department of Defense stalled. Anthropic objected to using its AI for mass surveillance of Americans and for autonomous targeting or firing decisions involving lethal weapons. The Pentagon argued that a private contractor should not dictate lawful military use of technology.

Government lawyers also said Anthropic might disable or alter models during wartime operations. Lin said she had not seen evidence that the company could change a delivered model or activate a remote kill switch. She also called the government's reliance on Anthropic's public criticism troubling because it could punish contractors for disagreeing with an agency.

The case reaches beyond one vendor. Frontier models are increasingly embedded in defense analysis, logistics and intelligence, while their developers maintain safety rules that may conflict with government demands. Agencies need predictable access, but suppliers also face responsibility for uses they believe their systems cannot perform safely.

The court is not deciding whether Anthropic's preferred military restrictions are correct. It is deciding whether the government followed a lawful and evidence-based process when imposing a severe designation. That procedural distinction protects agencies as well as contractors because emergency supply decisions become less reliable when policy disagreement is treated as a technical vulnerability.

The designation can affect more than one government contract. Suppliers, partners and foreign customers may interpret it as an official security judgment, making procedural fairness essential. The administration should identify specific technical risks and provide a path for remediation rather than relying on disagreement over model-use policy. Anthropic, meanwhile, must show that its restrictions can be implemented without disrupting legitimate defense work. A clear court record could help other AI vendors understand the evidence and process required before a company is excluded from a critical supply chain.

A supply-chain-risk label carries serious commercial consequences and should rest on technical evidence rather than a policy dispute. The court's decision may determine how much leverage the government has to force AI companies to remove use restrictions. It may also establish whether public criticism can become a factor in excluding a contractor from federal work.