Judge strikes down Pentagon blacklist of Anthropic as unlawful retaliation

U.S. District Judge Rita Lin ruled on Thursday that the Pentagon broke the law when it labelled Anthropic a supply-chain risk earlier this year. The 59-page order from the Northern District of California finds that the designation was retaliation for the company's refusal to loosen limits on how the military could use Claude, not a genuine finding that the software would sabotage U.S. systems.

Lin had paused the blacklist in March. The new decision makes that pause permanent for the federal agencies named in the suit. It also vacates Defense Secretary Pete Hegseth's directive that other Pentagon contractors stop doing business with Anthropic. The government is expected to appeal.

What the designation did

Anthropic's label was the first time a U.S. company had been publicly classed as a supply-chain risk under a procurement statute written to keep foreign sabotage out of military systems. Officials announced the step in February after President Donald Trump and Hegseth accused the firm of endangering national security.

The dispute began in late 2025, when the Defense Department asked Anthropic to put Claude on an internal platform called GenAI.mil. Officials wanted unrestricted use for what they called all lawful purposes. Anthropic chief executive Dario Amodei would not drop terms that blocked mass surveillance and autonomous armed drones. The department then imposed the risk designation and told contractors to cut ties.

Anthropic sued in March. It said the label could cost billions in lost contracts and damage its standing with civilian customers. Lin granted a temporary halt the same month, writing that the government's conduct looked like an attempt to cripple the company for exercising First Amendment rights.

What the court found

The Thursday order holds that the measures violated the First Amendment, the Fifth Amendment's due process clause, and the Administrative Procedure Act. Lin wrote that the empty invocation of national security is not a blank check to punish and retaliate against government critics.

She accepted that the Department of War can choose its own artificial intelligence vendor. The problem, she said, was the breadth of the penalty and the record behind it. The evidence showed the measures were illegal and baseless. Officials had continued to pursue work with Anthropic even after calling it a saboteur. That conduct, she wrote, is not consistent with a genuine fear that the company would poison its software.

The Pentagon had also claimed Anthropic retained a backdoor into models once they sat on a national security system. During the case, U.S. officials conceded that the company has no such access. Lin recorded the concession: Anthropic's technology is no riskier than any other black box model.

The order states that the government's actions were based on a desire to make a public example of Anthropic for what officials called arrogance in criticising the administration, not on an articulable basis to believe the firm would sabotage its model. Neither the Constitution nor the statute the government cited, she wrote, allows sweeping penalties based principally on a critique of official views.

What happens next

Lin's injunction bars the named agencies from enforcing Trump's order to stop using Anthropic tools. It does not force the Pentagon to buy Claude. The department remains free to pick another vendor. Anthropic still has a parallel complaint before the U.S. Court of Appeals for the D.C. Circuit. That court must also rule before the ban is fully lifted across the federal system.

An Anthropic spokesperson said the company welcomed the finding that the designation was unlawful and remained focused on working with the government on national security uses. The Defense Department did not immediately issue a detailed public reply. Officials have argued throughout that the military must have unrestricted access to frontier models.

The case now sits at the junction of two fights that will outlast this order: how far a contractor may refuse a use case the Pentagon wants, and how far an administration may go when it treats that refusal as a security threat. Lin's answer is narrow and specific. The government can walk away from a vendor. It cannot, on this record, brand an American company a supply-chain risk in order to punish speech.

Contractors that had already stepped back from Anthropic now have a court order saying the instruction to do so was void. Whether they return, and on what terms, will be a commercial decision. The legal question Lin answered is simpler. A national security label still needs facts. On Thursday the court said the facts were not there.