A Judge Has Stopped the Pentagon from Blacklisting Anthropic

A federal judge on Thursday prohibited the Trump administration from designating Anthropic as a national security concern related to its AI models, determining that such a designation was unconstitutional retaliation.

The ruling delivered by US District Judge Rita Lin in California overturned the February 27 decision by Defense Secretary Pete Hegseth, which labeled Anthropic a “supply-chain risk” and rendered the company ineligible for federal contracts. The ruling also rescinded an additional punitive measure imposed by Hegseth that barred contractors or suppliers to the US military from engaging in business with Anthropic, a decision Lin described as “arbitrary, capricious, an abuse of discretion, and otherwise not in accordance with law.”

“While the Department of War has the unquestionable right to choose its AI vendor, the evidence shows that the sweeping measures taken against Anthropic were unlawful and unfounded,” Lin stated in her 59-page judgment.

The judge further determined that nine agencies—including the Pentagon, the Treasury Department, the State Department, and the Department of Homeland Security—had unjustly imposed sanctions on Anthropic. This ruling eliminates those penalties.

However, Lin confirmed that the Pentagon is not obliged to utilize Anthropic’s models, and the ruling does not prevent the Pentagon from opting for alternative models. A spokesperson from the Pentagon could not be reached for immediate comment, but the department is anticipated to appeal.

In a statement, Anthropic spokesperson Danielle Cohen expressed, “We welcome the court’s decision that this supply-chain risk designation was unlawful. We remain committed to working collaboratively with the government to leverage AI for our national security.”

The case arises from a contentious dispute earlier this year between the Pentagon and Anthropic regarding a $200 million agreement to employ the AI lab’s Claude models for military purposes.

The disagreement emerged following reports that the US had employed Claude in the operation to apprehend Venezuelan President Nicolás Maduro. Subsequently, a Palantir employee conveyed concerns from an Anthropic employee to US officials about the application of its models.

During negotiations, Anthropic insisted on defining limitations on the deployment of its models, including restrictions against support for lethal autonomous weapons and mass surveillance systems. However, Hegseth dismissed any restrictions, arguing that a contractor could not dictate the use of technology once it was deployed and asserting the contract encompassed “all lawful use.”

After talks collapsed in February, Hegseth was broadly perceived as retaliating against Anthropic by designating it a “supply-chain risk,” a national security designation that effectively barred the company from federal contracts.

At that time, the Pentagon indicated that the supply-chain risk classification was justified, as granting Anthropic access to any classified systems would “introduce unacceptable risk” if the AI lab could disable or alter its technology during conflicts.

In reaction, Anthropic filed two lawsuits—one in federal district court in California and the other at the US Court of Appeals for the District of Columbia—claiming the Pentagon violated its First and Fifth Amendment rights on ideological grounds. The DC case is still pending.

Anthropic’s leading AI models are regarded as among the best globally. Apart from the supply chain risk classification, Anthropic’s models have recently fallen under the Trump administration’s AI oversight framework due to their advanced capabilities.

Lin noted the administration’s engagement in evaluating advanced Anthropic models as part of her ruling. “Even now, the government is exploring collaboration with Anthropic on its new model, Mythos, in various sensitive contexts,” she stated. “None of this aligns with a legitimate fear that Anthropic is a saboteur intending to compromise its software and jeopardize national security.”

Additional reporting by Maxwell Zeff.