Anthropic’s fight against the Department of Defense raises a big question: Who controls AI in war?

A federal judge has handed Anthropic a major victory in its escalating confrontation with the Pentagon, blocking an attempt to label the AI company a national security supply-chain risk after it refused to remove restrictions on how its Claude chatbot could be used by the U.S. military.

The dispute has developed into one of the most significant tests yet of how much control an AI company can retain over its technology once the government becomes a customer.

At the centre of the fight is a question likely to grow more important as artificial intelligence moves deeper into military planning and operations: Can the Pentagon dictate how a private AI company’s technology is used, or can the company draw its own red lines?


U.S. District Judge Rita Lin sided, at least for now, with Anthropic.

In a 59-page order issued Thursday, Lin blocked the Pentagon’s designation of Anthropic as a supply-chain risk, describing the action as “illegal and baseless” and rejecting the government’s attempt to justify the move on national security grounds.

The clash began over Claude’s military use

The fight traces back to Anthropic’s restrictions on the use of Claude, its artificial intelligence model.

Anthropic was willing to work with the U.S. military but opposed allowing its AI to be used for certain activities, including domestic surveillance and autonomous weapons.

The company has argued that today’s AI systems are not reliable enough to be trusted with decisions involving the use of lethal force without meaningful human control. It has also raised concerns that using AI for surveillance inside the United States could threaten civil liberties.

The Pentagon took a sharply different view.

According to the government, a private company should not be able to impose restrictions that interfere with military operations or create uncertainty about whether an AI system could suddenly become unavailable during a mission.

The disagreement eventually moved beyond contract negotiations and into a much more serious confrontation.

Pentagon’s blacklist put Anthropic in uncharted territory

Defense Secretary Pete Hegseth designated Anthropic a national security supply-chain risk, a rarely used label intended to protect U.S. military systems from potential infiltration, sabotage or other security threats.

The designation was extraordinary because it was the first known instance of a U.S. company being publicly targeted under the government procurement authority used in the case.

For Anthropic, the consequences could have been substantial.

The designation restricted the company’s access to certain Pentagon contracts and, according to Anthropic, threatened billions of dollars in future business while potentially damaging its reputation across the broader government sector.

Anthropic argued that the label had little to do with an actual security threat.

Instead, the company alleged that the Pentagon was retaliating against it for expressing views on AI safety and refusing to accept conditions that would have removed its restrictions.

Judge says national security is not a ‘blank check’

Judge Lin’s ruling delivered the clearest legal rebuke yet to the Pentagon’s handling of the dispute.

She found that the government could not simply invoke national security to punish a company over a disagreement about its position.

“The empty invocation of national security is not a blank check to punish and retaliate against government critics,” Lin wrote.

Anthropic’s lawsuit, filed in federal court in California on March 9, accused the government of violating its First Amendment rights by punishing the company for its views on AI safety.

The company also argued that it had been denied due process because it was not given a meaningful opportunity to challenge the designation before the government acted.

Anthropic further pointed to what it described as a contradiction in the Pentagon’s position, noting that the military had previously praised Claude and worked with the company before the dispute over restrictions intensified.

The Justice Department, however, maintained that the case was not about Anthropic’s opinions.

Government lawyers argued that the supply-chain designation resulted from the company’s refusal to agree to contractual terms and the operational risks that could arise from uncertainty over how Claude could be deployed.

Who gets the final say over AI in war?

The immediate dispute concerns Anthropic and the Pentagon, but the implications could reach far beyond Claude.

AI companies are increasingly being drawn into government and defence work as military organisations look for ways to use large language models and other advanced systems for intelligence analysis, logistics, planning and decision-making.

That expansion is creating a new fault line between the companies building the technology and the governments seeking to deploy it.

For years, technology companies have generally had more control over the limits they place on their products. Military contracts could test how long that remains possible.

If AI developers are allowed to set strict boundaries on how their systems can be used, they may be able to prevent their models from being deployed for autonomous weapons, mass surveillance or other applications they consider unsafe.

If the government can use its contracting power or national security authorities to override those boundaries, AI companies may face pressure to give the military far greater control.

That makes the Anthropic case about more than one chatbot or one government contract.

It is increasingly becoming a test of who sets the rules when AI moves from the technology sector into national security.

Anthropic still faces another legal fight

Thursday’s ruling does not end Anthropic’s broader conflict with the U.S. government.

The company is also pursuing a separate lawsuit in Washington, D.C., challenging another Pentagon supply-chain risk designation that could potentially affect its eligibility for civilian government contracts.

That case could determine whether the latest ruling marks an isolated setback for the Pentagon or the beginning of a broader legal limit on how the government can use national security powers against AI companies.

For now, however, Anthropic has secured an important victory.

The company built Claude with safety restrictions that it has repeatedly said are necessary to prevent AI from being used in ways that could cause serious harm.

The Pentagon argues those restrictions cannot be allowed to dictate the limits of U.S. military operations.

A federal judge has now stepped into that conflict — and her ruling could become an early marker in a much larger battle over artificial intelligence, autonomous weapons and the future of warfare.

As AI systems become more capable and more deeply embedded in military infrastructure, the question raised by Anthropic’s clash with the Pentagon is unlikely to disappear: when an AI company says no, who ultimately gets to decide?

(With agency inputs)