Record ‘has gotten worse for the government’ in Anthropic dispute, judge says

A federal district judge appears poised to again side with Anthropic in a legal challenge over the Trump administration’s decision to designate the company as a supply chain risk and enact a governmentwide ban on the technology. 

During a Thursday hearing, Judge Rita Lin of the U.S. District Court for the Northern District of California said the record in the dispute appears mostly the same as what she reviewed before issuing a preliminary injunction against the government. 

That March decision blocked the government from carrying out its agency-wide ban against the Claude maker and any Department of Defense actions to effectuate its supply-chain risk designation.

The record — which included a batch of emails between Anthropic’s Dario Amodei and DOD’s research and engineering leader, Emil Michael, leading up to the clash — also didn’t provide additional justification from the government. 

“If anything, it seems like the record, in some ways, has gotten worse for the government,” Lin said.

Lin said she did not see evidence that Anthropic could alter the model after it provided access to the government, that it could turn it off with a “kill switch,” or that the model was provided to DOD with nefarious controls. Rather, she said, the record suggests that Anthropic publicly criticized the DOD and that led to the government’s stated lack of trust in the AI provider.

As such, Lin said the government’s position appears to be that if a defense contractor criticizes the administration, that criticism can itself amount to a breach of trust and open the door for retaliatory actions, including taking away contracts and calling that company an enemy of the state. 

“If that’s really what the government’s position is,” Lin said she finds it “to be really troubling.”

The hearing on competing motions for the court to rule in each party’s favor is the latest in the legal battle at the heart of tensions between the Trump administration and Anthropic. 

The dispute began early this year when Anthropic went public with the details of its disagreement with the Pentagon over its technology’s use. In response, President Donald Trump and Defense Secretary Pete Hegseth issued directives via social media banning federal agency use of the technology and announcing the designation of the company as a supply-chain risk. 

Those actions had immediate impact, with agencies across the government halting use of Claude internally. Anthropic challenged the government’s actions in both district and circuit court.

In the interim, Lin’s preliminary injunction issued a temporary block allowing agencies to continue using Anthropic tools while the case moves forward. While not directly related to the legal challenge, the Trump administration also took the eyebrow-raising step of enacting export controls. 

During the Thursday hearing, Lin asked about the status of Anthropic use across the government and whether any agencies were winding down or exploring new uses of the company’s products.

James Harlow, a Justice Department lawyer representing the government, told Lin that the DOD continues its work to phase out the systems with a completion date of Sept. 30. And another unnamed defendant agency had preexisting plans to transition away from Anthropic and to another AI provider.

Harlow explained that elsewhere in government, use of Anthropic is nuanced in that some agencies are experimenting with the tools via sandboxes, pilot programs and through third-party vendors. He specifically referenced a pilot program several agencies are using that will end at the end of August. 

Unless agencies decide to continue or modify their use, after the Aug. 30 deadline, that use would stop, Harlow said.

While he didn’t name the program, it’s almost certainly the General Services Administration’s OneGov deals, through which several AI technology providers have offered their products at deeply discounted rates for roughly a year.

Meanwhile, Michael Mongan, a lawyer with Wilmer Hale representing Anthropic, said two unnamed agencies paused negotiations on new contracts, while other federal agencies are still pursuing new deals.

He said Anthropic is not aware of agencies canceling contracts since the preliminary injunction, though there was a contract with the Pentagon’s Chief Digital and Artificial Intelligence Office (CDAO) that was canceled before the court’s injunctive relief. As for private contractors winding down Anthropic use, Mongan said it’s hard for them to tell and much of the information is secondhand.

Mongan said they have heard secondhand reports about DOD communicating with contractors and subcontractors about Claude, but the government was in a better position to answer those questions.

Notably, a separate flare-up between the administration and Anthropic has taken place in the background while the litigation advanced.

In June, the Trump administration restricted Anthropic’s Mythos 5 and Fable 5 systems, citing national security concerns posed by a potential bypass of their security restrictions. But the export controls it put in place were lifted just a few weeks later after the administration and Anthropic worked out guardrails. 

Federal government use of Mythos came up at the Thursday hearing. 

Among her questions, Lin asked the government to provide details about expanded use of Anthropic tools in national security, including deployment of Mythos. Lin said that expanded use of Anthropic’s services, like Mythos, would be inconsistent with DOD’s arguments in the case.

Reports suggest the government is using the model.

Harlow initially argued that the question solicited nonpublic information and was irrelevant to the case, as the action would have come after the preliminary injunction. He later indicated it may be possible to submit something after the hearing, however.

Mongan said he agreed such a question about expanded use would be relevant but understood national security concerns from the government. He suggested a private discussion or brief explanation from the government post-hearing, but said Anthropic didn’t want the question to delay judgement.