A Judge Tells the Pentagon It Cannot Blacklist Anthropic

A federal judge has told the Pentagon it cannot punish Anthropic for refusing to let its AI be used for everything the military wanted. In a ruling handed down on Thursday, U.S. District Judge Rita Lin found that the Department of Defense broke the law when it branded the company a supply-chain risk, and she ordered the designation set aside.

The label was not a small administrative matter. A supply-chain risk designation can freeze a company out of federal contracts and steer other agencies away from its products. Lin concluded the government reached for it not because Anthropic posed a genuine security threat, but because the company had drawn public lines the Pentagon did not like.

What the fight was about

The dispute began earlier this year, when the Defense Department pushed to use Claude, Anthropic's flagship model, for "all lawful purposes," including the most sensitive military and intelligence work. Anthropic agreed to most of that request but kept two categories off the table. It would not allow its models to be used for mass surveillance of Americans, and it would not allow fully autonomous weapons that could take a human life without a person making the decision.

The military treated those limits as an obstacle. Anthropic treated them as a condition of doing business at all. Rather than settle the disagreement quietly, Lin found, officials moved to make a public example of the company. In her written opinion she was blunt about it.

"The empty invocation of national security is not a blank check to punish and retaliate against government critics."

Lin described the designation as "illegal and baseless" and "arbitrary and capricious," a legal test that asks whether an agency had any reasoned basis for what it did. She found that Anthropic's public criticism of the government was a substantial factor in the decision, which put the action on a collision course with the First Amendment.

Why it matters beyond one company

The ruling lands on a question the industry has mostly avoided in public. AI firms sell general-purpose systems, then try to write rules about how those systems may be used. Governments, and militaries in particular, want the tools without the conditions. Judge Lin's opinion says the government cannot use its procurement power as a cudgel to strip those conditions away, at least not by dressing up retaliation as a security concern.

Anthropic has spent much of the past year positioning itself as the safety-minded lab, and its refusal to hand the Pentagon a blank cheque fits that story. The company had earlier built a government-focused version of Claude for federal use, so this was never a refusal to work with Washington. It was a refusal to work with Washington on any terms at all.

The government is expected to appeal, so the fight is not over. For now, though, a court has said that an AI company is allowed to say no, and to say so out loud, without being punished for it.