A Judge Has Blocked the Pentagon’s Attempt to Blacklist Anthropic

A federal judge on Thursday barred the Trump administration from designating Anthropic a national security risk over a dispute about the use of its AI models, ruling that the designation amounted to unconstitutional retaliation.

The decision handed down by US district judge Rita Lin in California vacated the February 27 decision by defense secretary Pete Hegseth to label Anthropic a “supply-chain risk” and make the company ineligible for federal contracts. The ruling also lifted an additional punitive measure by Hegseth that prevented contractors or suppliers to the US military from doing business with Anthropic, a position Lin called “arbitrary, capricious, an abuse of discretion, and otherwise not in accordance with law.”

“Though the Department of War is undisputedly free to select the AI vendor of its choice, the evidence demonstrates that the broad measures imposed on Anthropic were illegal and baseless,” Lin wrote in a 59-page ruling.

The judge also ruled that nine agencies—including the Pentagon, the Treasury Department, the State Department, and the Department of Homeland Security—had improperly imposed sanctions on Anthropic. The ruling removes those penalties.

However, Lin affirmed that the Pentagon is not required to use Anthropic’s models, and the ruling does not prevent the Pentagon from electing to simply use other models. A Pentagon spokesperson could not immediately be reached for comment, but the department is expected to appeal.

In a statement, Anthropic spokesperson Danielle Cohen said, “We welcome the court’s ruling that this supply-chain risk designation was unlawful. We remain focused on working productively with the government to harness AI for our national security.”

The case stems from a bitter dispute earlier this year between the Pentagon and Anthropic over a $200 million deal to use the AI lab’s Claude models for military applications.

The rift began after reports that the US had used Claude in the operation to capture Venezuelan president Nicolás Maduro. After the operation, a Palantir employee relayed concerns from an Anthropic staffer to US officials about how its models had been used.

In negotiations, Anthropic insisted on certain limits on how its models could be deployed, including support for lethal autonomous weapons and mass surveillance systems. But Hegseth rejected any curbs, contending that a contractor could not dictate how the technology could be used once deployed and insisting the contract allowed “all lawful use.”

After negotiations broke down in February, Hegseth was widely seen as punishing Anthropic by designating it a “supply-chain risk,” a national security tool that blacklisted the company from doing business with the federal government.

The Pentagon said at the time that the supply-chain risk label was appropriate, because giving Anthropic access to any classified systems would “introduce unacceptable risk” if the AI lab could disable or alter its technology, for instance, at a time of war.

Anthropic filed two lawsuits in response—one in federal district court in California and one at the US Court of Appeals for the District of Columbia—accusing the Pentagon of violating its First and Fifth Amendment protections on ideological grounds. The DC case is ongoing.

Anthropic’s top AI models are considered some of the best in the world. Separate from the supply chain risk designation, Anthropic’s models recently became subject to the Trump administration’s AI oversight framework due to their powerful capabilities.

Lin cited the administration’s involvement in reviewing advanced Anthropic models as part of her ruling. “Even now, the government is discussing collaboration with Anthropic on its new model, Mythos, in an array of sensitive contexts,” she wrote. “None of that is consistent with a genuine fear that Anthropic is a saboteur who would poison its software to harm national security.”

Additional reporting by Maxwell Zeff.