The Ruling: "Illegal and Baseless"
On August 28, 2026, U.S. District Judge Rita F. Lin issued a 59-page ruling finding that the Trump administration's treatment of Anthropic constituted "unlawful retaliation" in violation of the First and Fifth Amendments. The government's conduct was, in her words, "illegal and baseless." That outcome capped a six-month legal fight that began with a single question: can an AI company be punished for writing a critical essay?
The Contract, and the Line Anthropic Would Not Cross
In July 2025, Anthropic signed a $200 million contract to integrate Claude across the Pentagon's classified networks, making it the first AI model deployed on those systems. The relationship soured that autumn during negotiations over a new platform called GenAI.mil, which was intended to give military personnel, civilians, and contractors access to frontier AI tools. The Pentagon asked Anthropic to accept language permitting Claude to be used for "any lawful use." Anthropic largely agreed, but drew two firm lines it would not cross: Claude could not be used for mass surveillance of Americans, and it could not be used to make final lethal decisions in fully autonomous weapons systems. Both conditions had been part of Anthropic's terms of service since the company's founding in 2021, and had governed the Pentagon relationship without incident for months.
The Ultimatum: Accept "Any Lawful Use" or Face the Consequences
Defense Secretary Pete Hegseth set a deadline of February 27, 2026. Accept the Pentagon's terms or be removed from its systems. CEO Dario Amodei rejected the ultimatum, writing publicly: "We cannot in good conscience accede to their request." Amodei had already signalled his position in a January 2026 essay flagging the risks of AI-powered surveillance and autonomous weapons. The Pentagon cited that essay and subsequent media interviews as evidence of "increasingly hostile manners." In other words: criticising the government's AI ambitions was treated as a national security threat. On February 27, President Trump directed federal agencies to cease using Anthropic's products. The same day, Hegseth designated Anthropic a "supply chain risk." That designation, which bans defence contractors from using the labelled vendor, had never previously been applied to an American company. It is the label normally reserved for companies considered to be connected to foreign adversaries.
"A Desire to Make a Public Example"
Anthropic filed suit on March 9, 2026. Seventeen days later, Judge Lin issued a preliminary injunction, calling the Pentagon's move "classic illegal First Amendment retaliation" and invoking a word federal judges use sparingly: Orwellian. In her August 28 final ruling, Lin went further. She found the ban was "based on a desire to make a public example out of Anthropic for its arrogance in criticising the government." The court dismissed both of the Pentagon's core legal arguments.
The Pentagon's Arguments, and Why Both Failed
The first argument: Anthropic had hidden backdoor access to deployed systems. That claim was acknowledged as unfounded by government officials during proceedings. The second: Anthropic had shown "increasingly hostile conduct" by criticising the Administration in the press. Lin addressed it directly: "Neither the Constitution nor the federal statute invoked by Defendants allows them to impose sweeping penalties based principally on Anthropic's critique of the Administration's views." Internal evidence undercut the Pentagon's stated fears. Other parts of the government continued meeting with Anthropic after the ban, and the Defence Department itself had been using Claude since March. "None of that," Lin wrote, "is consistent with a genuine fear that Anthropic is a saboteur who would poison its software to harm national security."
"The empty invocation of national security is not a blank check to punish and retaliate against government critics.
"Steph30th of August 2026
OpenAI Stepped In Within Hours
While the case was working through the courts, the Pentagon did not go without an AI vendor. Within hours of Anthropic's February 27 rejection, OpenAI signed its own agreement with the Department of Defence. OpenAI had previously restricted weapons applications in its own terms of service; those restrictions were quietly removed ahead of the deal. The contrast sharpened the question at the centre of the case. Two frontier AI companies faced the same choice. One refused and got sued. The other agreed and got the contract. Judge Lin's ruling does not change what OpenAI chose, but it does establish that the government could not lawfully force Anthropic to make the same call.
What Comes Next: the Pentagon Will Fight It
The ruling is a win for Anthropic, but the case is not closed. The Defence Department filed a notice of appeal, bringing Judge Lin's order to the Ninth Circuit Court of Appeals. A separate Anthropic lawsuit in the Washington, D.C. Circuit, involving a different Pentagon rule, is also still pending. The government must also resolve a contradiction the ruling exposed. Publicly, the Pentagon argued Anthropic was a national security threat. Internally, it kept using the product. The administration will have to decide whether to keep fighting a case whose own evidence undercuts its position, or find a different legal path to the "any lawful use" language it wants.
The Precedent That Outlasts This Case
A federal court has now established that writing a critical essay about your government contract is protected speech, and that a supply chain risk designation deployed to punish that speech is unconstitutional retaliation. Any AI company, defence contractor, or vendor doing business with the federal government can point to that ruling the next time it is asked to sign away its own terms of service. That is the consequence that outlasts the appeal, the Ninth Circuit, and whatever the two parties eventually settle on.