Judge Pushes Back Hard on Pentagon's Case Against Anthropic

A federal judge told the Trump administration on Thursday that it still hasn't backed up its claims against Anthropic, dealing another setback to the government's effort to bar the AI company's technology from federal use.

U.S. District Judge Rita Lin, presiding over the case in California, said the administration lacks sufficient evidence to justify labeling Anthropic a "supply chain risk," the designation at the heart of a monthslong legal fight between the AI maker and the Department of Defense. Bloomberg and Axios first reported details of the hearing.

The clash traces back to contract negotiations that broke down between Anthropic and the Pentagon. Anthropic has said it doesn't want its AI systems used for mass surveillance of Americans or folded into decisions about targeting and firing lethal weapons, telling the government the technology simply isn't ready for that kind of use. The Pentagon pushed back on that stance, arguing that a private vendor shouldn't get to set limits on how the military deploys its tools, and insisting it would only use Anthropic's systems in "lawful" ways.

A shaky legal theory

Things got more contentious from there. The government has argued, in part, that Anthropic's public criticism of the Department of Defense is itself grounds for the supply chain risk designation and the resulting ban. Judge Lin didn't let that argument slide. She called the reasoning "really troubling," and warned that accepting it could open the door to retaliation against any federal contractor that publicly disagrees with the administration — a precedent with implications well beyond one AI company.

The Pentagon also raised a more technical claim: that Anthropic could potentially disable or alter its models mid-deployment during military operations, including in the middle of warfighting. That's a serious allegation if true, since it would suggest a contractor could unilaterally sabotage systems the military depends on in the field. But according to the reporting, experts say there's no evidence to support it, and Lin said she'd seen nothing showing Anthropic has the ability to modify a model once it's been delivered, or to trigger what she described as "some kind of kill switch."

What's actually being decided

Thursday's hearing is part of one of two lawsuits Anthropic filed against the Department of Defense back in March, both challenging the supply chain risk label and the resulting ban on the company's technology in federal use. This case is playing out in California in front of Judge Lin; a second is being heard separately in Washington.

Lin already stepped in once, issuing a temporary block on the ban back in March. Now she's weighing whether to convert that temporary order into a permanent one — a decision that would settle, at least for this case, whether the government can keep Anthropic's technology out of federal contracts based on the evidence presented so far.

The stakes go beyond a single vendor dispute. Anthropic is one of the major players building the large language models that federal agencies are increasingly looking to adopt, and a permanent ban would cut the company out of a significant customer. But the broader question hanging over the case is what kind of leverage the government has over AI companies that push back on how their tools get used, especially in military contexts. Judge Lin's comments suggest she's wary of letting public criticism of the Pentagon count as justification for locking a contractor out of federal work, a standard that could reshape how AI companies are willing to speak up about the terms of their government deals.

For now, the ban Lin blocked in March remains on hold while she decides whether to make that pause permanent.