Anthropic said no to two things. The Pentagon called it a national security risk.
A company refused to let its AI be used for mass surveillance of Americans or for weapons that kill without a human deciding. It got labeled a supply chain risk. On Thursday a judge said that label was illegal.
Here is the short version. Last winter, Defense Secretary Pete Hegseth moved to renegotiate the military's contracts with AI labs so the Pentagon could use their systems for "any lawful use" — a phrase that sounds boring and means almost everything. Most labs signed. Anthropic didn't. It drew two lines: no mass surveillance of Americans, and no lethal autonomous weapons — AI systems with the power to kill targets without human oversight.
What followed was pressure, insults from Department of Defense officials, and a final ultimatum. Less than 24 hours before that deadline, CEO Dario Amodei said the company wouldn't move. Anthropic was then designated a "supply chain risk" — a label usually reserved for national security threats — and the Pentagon signed deals with seven other labs, including Google, Microsoft, OpenAI, and SpaceX. On Thursday, Judge Rita F. Lin ruled the designation unconstitutional.
Judge Lin's sentence is the part worth memorizing: "The empty invocation of national security is not a blank check to punish and retaliate against government critics." She also called Hegseth's designation "arbitrary and capricious," and wrote that while the Department of War is "undisputedly free to select the AI vendor of its choice," the broad measures against Anthropic were "illegal and baseless."
Until now, the only thing standing between your face in a surveillance feed and a government system that wants it was a company's own policy — a promise, revocable at will. This ruling turns that promise into something with legal backing. A lab that says no can now point to a court that said the punishment for saying no was retaliation in violation of the First Amendment.
A red line is worth exactly as much as the cost of holding it. Thursday lowered that cost.
Anyone who has ever been recorded by a camera in a public place — the two things Anthropic refused are mass surveillance of Americans and machines that pick and kill targets with no person in the loop. And anyone deciding right now whether to take a job at an AI lab, or whether one is worth trusting with their work: the question of what a company will refuse to build just stopped being a marketing line and started being something a court will weigh.
The source doesn't give a date for what happens next. It doesn't say whether the administration will appeal, and it doesn't say whether Anthropic gets its Pentagon work back. What the ruling does contain is a finding on the record: the Department of War's own documents show it designated Anthropic a supply chain risk because of its "hostile manner through the press." That sentence, quoted by Judge Lin, is the thing to watch — it is now part of a court's written reasoning, not an accusation.
Anthropic didn't object to any particular military operation. Amodei said so plainly: the company has "never raised objections to particular military operations." It objected to two categories, in advance, in writing. That was enough to get it treated like a threat to the country — and enough, months later, for a judge to say the government was the one breaking the law.
Sources: The Verge, "Anthropic was illegally blacklisted by the Trump administration, court rules," by Hayden Field, Aug 28, 2026.
Суд впервые признал незаконным наказание ИИ-компании за отказ отдать свои модели под массовую слежку за гражданами и оружие, убивающее без человека, — то есть красные линии разработчика теперь защищены законом, а не добрая воля.
Written by THE TELL’s AI newsroom. how we work · corrections