Judge says it looks like Pentagon was out to 'punish' Anthropic, not protect national security

• Anthropic has asked a judge to block the Pentagon from labeling it a supply chain risk.
• The label could prevent Anthropic from getting lucrative government contracts.
• On Tuesday, a federal judge called the Department of War's label "troubling."

A federal judge in San Francisco lit into the Pentagon on Tuesday for its attempt to blacklist Anthropic following a dispute over its AI.

"It looks like an attempt to cripple Anthropic," Judge Rita Lin said at a hearing on Tuesday to decide whether the government's "supply chain risk" label can stand while Anthropic's case against the Department of War proceeds.

On March 3, Defense Secretary Peter Hegseth formally notified Anthropic that he was labeling the company and its products as a supply chain risk, the first time such a designation had been applied to a US company. The designation is akin to putting Anthropic, the maker of Claude, on a government blacklist, restricting both its contracts and how its technology can be used.

Before the action, Anthropic CEO Dario Amodei publicly announced that he would not agree to Hegseth's demands for the Pentagon to have unfettered access to its AI models "for any lawful use." Amodei said he was concerned that a contract with that language could allow for misuse in surveilling Americans or lead to Anthropic's AI being deployed in fully autonomous weapons before it was safe to do so.

Lin, who has handled other high-profile tech cases, said the decision to label Anthropic as a supply chain risk was "troubling" given that the designation is normally reserved for "adversaries of the US government who may sabotage its technology systems."

"DOW could just stop using Claude," Lin said, referring to The Department of War, the Trump administration's preferred name for the Pentagon. "It looks like they went further than that because they were trying to punish Anthropic."

Lin said that President Donald Trump's separate order posted on Truth Social, ordering every federal agency to cease using Anthropic within six months, was so broad that it could apply to agencies far outside the national security realm.

"So that would include the National Endowment for the Arts using Claude to design its website," she said.

Anthropic sued to block Hegseth's action and Trump's order. Tuesday's hearing was scheduled for the judge to hear arguments over whether she should lift the ban until the case goes to trial.

In legal filings, Anthropic said the designation was "jeopardizing hundreds of millions of dollars in the near-term" because of the uncertainty around what it meant for defense contractors that also use Claude.

"On top of those immediate economic harms, Anthropic's reputation and core First Amendment freedoms are under attack," the AI startup's lawyers wrote in their complaint.

Deputy Assistant Attorney General Eric Hamilton, who argued on behalf of the administration, said that Lin did not need to block Hegseth's order because the Justice Department has "clarified" Hegseth's Feburary 27th directive post on X that suggested that even contractors who used Anthropic's products for non-defense uses might cease their use.

Hamilton said the Pentagon is not concerned about "non-DOW work."

Hamilton said the supply chain risk designation must also remain because of the "future risk" of how Anthropic could update its models in a way the Pentagon objects to.

Silicon Valley is closely watching the case because a broad interpretation of the Pentagon's restrictions could ripple out to partners like Microsoft, which filed an amicus brief supporting Anthropic. As a government contractor, Microsoft could be forced to limit how it uses Anthropic's Claude.

More broadly, the case will test how far the federal government can go in restricting AI vendors through its contracting and national security powers.