A six-month fight lands its first real verdict

A federal judge has ruled that the Pentagon broke the law when it punished the AI company Anthropic for publicly disagreeing with the Defense Department's stance on how the military should be allowed to use artificial intelligence. In a 59-page written order issued Thursday night, U.S. District Judge Rita Lin found that the Department of Defense's decision to label Anthropic a "supply chain risk to national security" was retaliation for protected speech, not a legitimate security judgment. The government is expected to appeal.

The ruling caps roughly six months of escalating conflict between one of the country's leading AI labs and its own military, a dispute that has forced courts to weigh in on a question with few clear precedents: how much authority the executive branch has to punish a private company for the positions it takes in public.

How the standoff began

The fight traces back to February, when President Trump ordered every federal agency to immediately stop using Anthropic's technology, and Defense Secretary Pete Hegseth designated the company a supply chain risk, a label the government normally reserves for firms tied to adversarial nations like China or Russia. The trigger was a contract dispute. Anthropic had signed a Pentagon contract worth up to $200 million in mid-2025 to deploy its Claude AI models on classified military networks, but the company insisted on two conditions: Claude could not be used to power fully autonomous weapons systems that select and fire on targets without human authorization, and it could not be used for mass domestic surveillance of American citizens.

The Pentagon wanted those restrictions lifted, pushing Anthropic and several other AI companies to agree their tools could be used for "all lawful purposes" without company-imposed carve-outs. Anthropic refused. Trump and Hegseth responded by cutting the company off entirely, and rival OpenAI announced a competing Pentagon deal within hours of the announcement, adopting similar safety commitments but structured as voluntary internal policy rather than binding contract terms the government would have to honor.

What the judge actually found

Anthropic sued in March, arguing the supply chain risk label violated its First Amendment rights and exceeded the statutory authority the government invoked to apply it. Judge Lin had already granted a preliminary injunction back in March, calling the designation "Orwellian" and writing that "nothing in the governing statute supports the notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government." Thursday's ruling went further, delivering a fuller judgment on the merits rather than just a temporary pause.

Lin wrote that the government's actions "were based on a desire to make a public example out of Anthropic for its 'arrogance' in criticizing the government, not based on any articulable basis to believe that Anthropic would actually sabotage its model." She added that neither the Constitution nor the federal statute the government cited allows officials to "impose sweeping penalties based principally on Anthropic's critique of the Administration's views." At an earlier hearing in July, Lin had told government lawyers the administration's position seemed "at odds" with the First Amendment and said she believed "the record had gotten worse for the government" as the case proceeded.

The government's counterargument

The Pentagon has consistently framed the dispute differently, arguing that private companies should not be able to constrain how the U.S. military uses technology it has purchased, and that Anthropic's contractual restrictions created operational uncertainty during active military operations. In earlier court filings, Justice Department lawyers argued that AI models are "so staggeringly enormous and opaque" that the Defense Department cannot evaluate them the same way it would a physical piece of hardware, making the case that stricter oversight authority was justified. Government attorneys have also maintained that the supply chain risk designation stemmed from Anthropic's refusal to accept specific contract terms, not retaliation for its public statements about AI safety.

Under Secretary of War Emil Michael pushed back sharply on the earlier preliminary ruling in March, writing on social media that it contained "dozens of factual errors" and arguing the decision "seeks to upend the president's role as Commander in Chief" during an active military conflict, a reference to the ongoing Iran war. The administration has signaled it views the underlying supply chain risk designation as still technically in effect while it pursues an appeal, disputing exactly how far the court's order extends.

What's actually at stake beyond one company

The case matters well past Anthropic's own business prospects. Claude had been the only AI model operating inside the Pentagon's most sensitive classified networks before the dispute, work that reportedly extended to operations including the capture of a foreign head of state. Losing that access created real logistical strain, with defense officials privately telling reporters it would be a genuine headache to disentangle Claude from existing systems. In the months since the initial designation, the Pentagon has moved to build out a broader, multi-vendor AI stack instead, signing separate classified-network agreements with OpenAI, Google, xAI, and several other companies.

The bigger question the case raises is one that extends to any company doing business with the federal government: whether an administration can use national-security-flavored designations, tools historically reserved for genuine foreign threats, as leverage against domestic companies that decline to go along with a policy position. An Anthropic spokesperson said in a statement that the company welcomes Thursday's ruling and "remains focused on working productively with the government to harness AI for our national security so all Americans benefit from this technology."

Where the fight goes from here

Thursday's ruling is not the final word. The government has signaled it intends to appeal, and a separate, narrower case Anthropic filed in the federal appeals court in Washington, D.C. remains pending alongside the California proceeding. Whatever the appellate courts ultimately decide, the underlying tension driving this case, how far the executive branch can go in penalizing companies for the positions they take in public, particularly around emerging technology with genuine national security implications, is unlikely to be settled by one ruling. It's a dispute other AI companies, defense contractors, and federal agencies are all watching closely, since the legal standard this case eventually sets will shape how every future disagreement between Washington and its technology suppliers gets negotiated, or litigated.