A federal judge said Thursday that the Trump administration’s case against Anthropic had “gotten worse,” sharply questioning whether officials labeled the AI company a national security threat because it had publicly criticized the government.
“I don’t see additional evidence from the government really justifying what it did,” U.S. District Judge Rita F. Lin said at the start of a hearing in the lawsuit the company brought against the administration. “If anything, it seems like the record, in some ways, has gotten worse for the government.”
Anthropic sued after the Pentagon designated it a national security supply chain risk and the administration moved to cut it off from federal work. The company says officials retaliated after it resisted efforts to remove restrictions on using its Claude models for domestic surveillance of Americans and lethal autonomous warfare.
The lawsuit seeks to overturn the designation and prevent the administration from excluding the company from federal contracting. Lin made the comments during a 2.5-hour hearing Thursday in California, where she was considering motions from each side to decide the lawsuit in their favor at this early stage in the case.
Trending
Lin said she saw no evidence Anthropic could alter its AI models after delivery, flip a kill switch or seed them beforehand with a hidden control mechanism. She also said she saw no specific reason to suspect the company would do any of those things.
Instead, Lin said, the record appeared to show that Anthropic’s public criticism of the Defense Department caused the administration to lose trust in the company.
“I find that position, if that’s really what the government’s position is, to be really troubling,” Lin said. “It seems at odds to me with the First Amendment, with Section 3252,” the statute the administration relied on to designate the company a national security supply chain risk.
She later described the government’s apparent position as “quite extreme.”
Lin also questioned whether President Donald Trump had evidence that Anthropic posed a security threat when he first publicly targeted the company on Feb. 27.
“For me, looking at the record doesn’t look like as of February 27th, there’s any information in the record about what Anthropic could or couldn’t do, or whether it posed a threat,” Lin said.
The risk analysis cited by the government was dated March 2, after Trump and Defense Secretary Pete Hegseth issued their directives against Anthropic. The company has argued that a memo prepared later could not justify actions officials had already taken.
Lin questioned whether the government had shown any specific reason to believe Anthropic would place hidden restrictions in future models.
“What are the specific articulable facts that suggest Anthropic might sabotage its software in this way?” she asked.
Justice Department lawyer James Harlow pointed to several things, including Anthropic’s refusal to permit all lawful uses, an executive’s questions about Claude’s use during an overseas military operation and concerns about the company’s handling of sensitive information.
Anthropic’s attorneys said none of those showed an intent to interfere with military operations. The company also noted that it had offered contract language promising not to do so.
Anthropic lawyer Michael Mongan argued that the government had shifted from its previously stated concerns about interference with deployed models to the possibility of hidden restrictions inserted before delivery.
That “shifting of rationales” was “powerful evidence of pretext” and weakened the government’s defense against Anthropic’s First Amendment and Administrative Procedure Act claims, he said.
Lin previously issued a preliminary injunction blocking parts of the administration’s actions.
The Pentagon continues to wind down its use of Anthropic products and expects to finish by Sept. 30, Harlow told the court. Several pilot programs at other agencies are scheduled to expire Aug. 30.
Mongan said most agencies had continued working with Anthropic, two had paused negotiations for new contracts, and others were negotiating new deals. He said the company was not aware of any agency canceling a contract after the injunction.
The Justice Department declined to say in open court whether any agencies had increased their use of Claude for national security work since the injunction.
Lin said that conduct could shed light on the government’s original intent and that she may order the parties to provide more information.
Anthropic and the Justice Department have each asked Lin to resolve the case through summary judgment, arguing that there are no material factual disputes that require a trial.
Lin did not rule at Thursday’s hearing. She took the competing motions under submission and said she would issue a written order.
Author
