The Pentagon cast Anthropic as a national-security threat. A federal judge found it had instead turned that label into a weapon in a fight over how far military AI should be allowed to go.

The conflict began over the winter, when Defense Secretary Pete Hegseth sought contract terms allowing the military to use AI for “any lawful use.” Anthropic accepted military work but drew two hard lines: no mass surveillance of Americans and no lethal autonomous weapons. Most major AI labs accepted the broader terms; Anthropic did not.¹

The administration answered with escalating pressure. President Donald Trump and Hegseth designated the maker of Claude a “supply-chain risk,” directing federal agencies to stop using its products and barring defense contractors from doing business with it. The White House framed the dispute as a matter of military control, saying Trump would not let a “woke company” dictate how the armed forces operate.²

Anthropic sued in March, calling the blacklist retaliation for its public safety limits. Its central argument was not that the Pentagon must buy its software, but that the government could not punish a company for speaking out about the uses of its own technology. Judge Rita Lin initially froze the directives, then on Thursday delivered a broader ruling vacating the designation.

Lin found the government had no substantiated basis to treat Anthropic as a saboteur, noting that the Pentagon continued pursuing work with the company and collaborating on cybersecurity. “The empty invocation of national security is not a blank check to punish and retaliate against government critics,” she wrote.³

The judge also rejected the premise underlying the risk label: Anthropic, she said, undisputedly lacks backdoor access to its models once the Defense Department deploys them. Still, the administration has argued that enormous, opaque AI systems cannot be assessed like ordinary hardware, and it is expected to appeal.⁴

Anthropic welcomed the decision while striking a conciliatory note, saying it remained focused on working with government to use AI for national security. A separate challenge to another Pentagon designation remains pending in Washington, leaving the larger battle over military AI far from over.⁵