A federal judge thinks the Trump administration is playing politics with AI contracts and seeking to punish Anthropic rather than the company being a real security threat.
US District Judge Rita F. Lin told a court hearing that the Trump administration’s actions to ban Anthropic as a security threat “ look like an attempt to cripple Anthropic, and such action, of course, would be a violation of the First Amendment.
The hearing comes as Anthropic tries to block a federal ban on government use of its AI models. Lin has not ruled yet, but she opened by saying the government’s response looked like punishment after Anthropic aired its dispute with the Pentagon.
Lin said that after Anthropic disclosed the row, the administration “seems to have a pretty big reaction to that and its actions don’t seem to be really tailored to a stated national security concern.”
She added: “It looks like the defendant went further than that because they were trying to punish Anthropic.”
Anthropic sued this month after being labelled a supply-chain risk, a tag more commonly aimed at Chinese entities. The fight stems from disagreement over how its Claude models can and cannot be used in national security work.
Lin asked for more evidence before deciding whether to grant an injunction. Anthropic says the designation is government overreach, and the administration has failed to justify the move.
WilmerHale attorney Michael Mongan said, “This is something that has never been done with respect to an American company.”
He argued the supply-chain label is normally reserved for foreign adversaries posing serious security threats.
In late February, President Donald Trump directed federal agencies to stop working with the company. Defence Secretary Pete Hegseth posted that military contractors and suppliers must stop using Anthropic.
Anthropic says the ban has already cost it hundreds of millions of dollars in cancelled contracts and scrapped customer agreements. It projects billions of dollars in lost revenue this year, tightening the screws on fundraising.
Even with the courtroom drama, Anthropic’s models are reportedly being used in the war with Iran for targeting and planning airstrikes. The outcome could reshape how the Pentagon deals with an industry that only recently stopped pretending it was above defence work.
The government argues Anthropic overreached by trying to enforce hard limits on how the Pentagon uses Claude. Anthropic wanted assurances that the models would not be used in fully autonomous weapons or domestic surveillance.
The Pentagon countered that prohibitions were unnecessary because existing military policy or law already restricts those uses.
Justice Department deputy assistant attorney general Eric Hamilton conceded that the Defence Department did not follow protocol in applying the supply-chain designation.
That process includes briefing Congress and exploring less intrusive alternatives, Hamilton acknowledged. The administration says it acted because Anthropic could later change its models or disable the technology mid-operation.
“It’s because of the risk of future subversion or sabotage,” Hamilton said. Anthropic says that kind of sabotage is technically impossible because it cannot access models after deployment.
In filings, Anthropic disclosed an email from the Pentagon, undersecretary for research and engineering Emil Michael, dated 4 March, five days after the president and defence secretary posted that the government would not do business with Anthropic.
Michael told Anthropic, chief executive Dario Amodei, they “are very close here” to an agreement to keep Anthropic’s tech in the Pentagon.
“I hope this works as I am running out of time,” wrote Michael, and Anthropic notes the Pentagon letter confirming the supply-chain designation is dated 3 March. Anthropic says it was “inconceivable” that Michael’s negotiations “reflected genuine concerns that Anthropic would sabotage military operations.”
