Judge says Trump admin still lacks evidence for Anthropic ‘supply-chain risk’ label

A federal judge just told the Trump administration what the technical community has argued for months: the "supply-chain risk" designation slapped on Anthropic is a political cudgel, not a security finding. U.S. District Judge Rita Lin didn't mince words. She called the government's rationale "really troubling" — a rare judicial rebuke that signals the court sees retaliation, not risk management.

The ban originated in stalled contract talks between Anthropic and the Department of Defense. Anthropic drew a line: its models would not power mass surveillance of Americans or feed targeting algorithms for lethal weapons. The company argued the technology wasn't ready for those missions. The Pentagon responded that a private vendor has no business dictating military doctrine, insisting it would use the tools only in "lawful" ways. That framing — lawful by whose definition, enforced by what oversight — went unexamined in the government's filings.

When the contract impasse hardened, the administration reached for a heavier instrument. It designated Anthropic a supply-chain risk, effectively blacklisting the company from federal procurement. The justification? Anthropic had publicly criticized the Defense Department. Lin saw through that instantly. A precedent that punishes contractors for disagreeing with an administration turns procurement into a loyalty test. That should unsettle every company that sells to the federal government, not just AI vendors.

The Pentagon's backup claim was more theatrical than technical: Anthropic could supposedly disable or alter its delivered models mid-conflict, a "kill switch" scenario. Experts dismissed it as fantasy. Lin agreed. She saw no evidence Anthropic could mutate a model already handed over, no mechanism for remote sabotage. The government brought speculation to a hearing that demanded proof.

Anthropic filed two lawsuits in March challenging the ban and the risk label. Thursday's hearing covered one; the other proceeds in Washington. Lin already issued a temporary injunction blocking the ban. Now she weighs whether to make it permanent. Her skepticism during oral argument suggests she leans toward keeping the government's thumb off the scale.

The stakes exceed Anthropic's contract. If a supply-chain risk designation can be weaponized against a contractor for policy dissent, the designation loses all meaning. It becomes a quiet tool for enforcing ideological conformity across the defense industrial base. Lin understands this. Her questions pressed the government on exactly that slippery slope.

The administration's defenders will frame this as military necessity versus corporate obstinacy. That framing dodges the core issue: a risk label requires evidence of risk. Not disagreement. Not criticism. Not hypothetical sabotage scenarios that collapse under cross-examination. The judge demanded evidence. The administration produced none.

What happens next will clarify whether the federal procurement system still distinguishes between security judgments and political score-settling. Lin's decision, whenever it lands, will answer that question for every contractor watching.