Federal Judge Voids Pentagon's Blacklist on Anthropic

A federal judge ruled the Pentagon’s supply chain risk designation against Anthropic was unlawful retaliation, arriving weeks before the AI firm’s record public offering.

Key Takeaways

• U.S. District Judge Rita Lin found the Pentagon’s February designation of Anthropic as a national security risk violated the First and Fifth Amendments.
• The ruling lands weeks before Anthropic’s expected listing, which bankers are targeting near $1.5 trillion to $2 trillion after a $965 billion funding round in May.
• The government has signaled an appeal to the Ninth Circuit, while a related case over a separate Pentagon rule remains pending before the D.C. Circuit.

The Ruling

A federal judge in San Francisco has delivered Anthropic a resounding victory in its confrontation with the Pentagon, finding that the Department of Defense broke the law when it branded the artificial intelligence company a national security supply chain risk. In a 59-page order issued Thursday evening, U.S. District Judge Rita F. Lin concluded that the designation amounted to retaliation for Anthropic’s public criticism of the department’s approach to military AI, a violation of both the First Amendment’s protection of speech and the Fifth Amendment’s guarantee that no penalty of this magnitude arrives without a hearing.

The record, Lin wrote, showed that the government’s conduct constituted unlawful retaliation, and that Anthropic never received the process the Constitution requires before such a measure takes effect. At a July 30 hearing, government lawyers had argued that artificial intelligence systems are too vast and too opaque to be evaluated the way the Pentagon evaluates conventional hardware, an argument the judge described at the time as troubling. Her written opinion goes further still, concluding that the measures were driven by a wish to make an example of a contractor for its defiance rather than by any documented security finding, a distinction that will matter well beyond this single case.

How the Dispute Began

The confrontation traces to February, when Defense Secretary Pete Hegseth issued Anthropic an ultimatum: open Claude to unrestricted military use or risk a Pentagon contract then worth roughly $200 million. Anthropic’s resistance centered on two limits written into its usage policy, a ban on deploying Claude for mass domestic surveillance and a refusal to let the model operate fully autonomous lethal weapons. Chief executive Dario Amodei held the line publicly, stating that the company could not in good conscience accede to the request, even as he acknowledged the department’s right to choose contractors who share its terms. He noted, too, the peculiarity of the government’s position: one arm of the administration was labeling Anthropic a security threat while another was pressing agencies to treat Claude as indispensable to national defense.

Hegseth answered on March 3 by naming Anthropic a supply chain risk, the first time that designation, ordinarily reserved for firms tied to foreign adversaries, had been applied publicly to an American company. The fallout arrived quickly: the General Services Administration dropped Anthropic from its federal purchasing contract, the Lawrence Livermore National Laboratory cut its access to Claude, and President Trump directed federal agencies to stop using the company’s products altogether. The reversal was striking given that Anthropic had, by its own account, been the first AI developer allowed inside classified government networks and the national laboratories in the first place.

A Case in Two Courts

Anthropic sued within days, and by late March Lin had granted a preliminary injunction pausing both the designation and the federal usage ban while the case proceeded on its merits, a ruling the government promptly appealed to the Ninth Circuit even as litigation continued beneath it. That appeal remains open, and Thursday’s judgment resolves only the Northern District of California proceeding. A second and narrower case, filed the same week under separate statutory authority, sits before the D.C. Circuit Court of Appeals, where a three-judge panel heard arguments in May and has not yet ruled on a different Pentagon rule invoked to justify the same blacklisting.

The Justice Department has signaled that it intends to appeal Thursday’s decision as well, which means the central legal question, how far a federal agency may go in penalizing a contractor over its public policy positions, is likely to reach a higher court before the underlying dispute or Anthropic’s own corporate timeline is fully settled. Anthropic’s response has been notably restrained given the scale of the vindication: a company spokesperson said only that Anthropic welcomed the ruling and remained focused on working productively with the government on national security applications. The White House did not respond to requests for comment, and the Pentagon has given no indication of whether it will pursue the designation through a different mechanism while its appeal is pending.

The Market’s Read

The commercial stakes of this dispute were never abstract. Anthropic’s models sat inside Palantir’s Maven Smart System, the Pentagon’s principal intelligence and targeting platform, tying the standoff to more than $1 billion in defense contracts and forcing a months-long technical rebuild once the ban took hold. Palantir shares rose roughly four percent in the days after the designation as investors wagered that rival vendors would fill the resulting gap, while defense primes including Lockheed Martin, Northrop Grumman, General Dynamics and RTX posted smaller gains on similar reasoning, evidence that the market’s clearest read on this dispute has run through Anthropic’s counterparties rather than through Anthropic itself.

That indirection exists because Anthropic carries no public equity against which investors can register a verdict of their own. What Thursday’s ruling does shape is the valuation underpinning its approaching debut. The company closed a $65 billion funding round in May at a valuation of $965 billion and has since told investors that its annualized revenue run rate reached $65 billion by the end of July, up from $47 billion in May and $9 billion at the close of last year. Secondary marketplace data from Nasdaq Private Market pegged shares at roughly $704 apiece in mid-August, suggesting private buyers are already pricing in growth beyond the funding round. Bankers are said to be targeting a public valuation of $1.5 trillion to $2 trillion, a multiple near fifteen times the disclosed run rate, once the prospectus, filed confidentially with securities regulators in June, becomes public after Labor Day.

What the Verdict Does Not Resolve

The episode has set Anthropic apart from its principal competitors. OpenAI, Google and xAI each accepted looser Pentagon terms during the same period, leaving Anthropic the only frontier laboratory willing to litigate rather than concede its usage limits, a posture that has become part of the company’s own commercial identity even as it courts public shareholders for the first time. The dispute drew bipartisan notice in Washington as it unfolded. Senator Thom Tillis, a North Carolina Republican, questioned why a disagreement with a strategic vendor had been allowed to play out so publicly, while Senator Richard Blumenthal, a Connecticut Democrat, argued that the administration’s own stated rationale, Anthropic’s refusal to permit mass surveillance of Americans, was itself the more troubling fact. The underlying policy debate over surveillance and autonomous weapons remains far from resolved regardless of Thursday’s outcome.

For a company on the cusp of what bankers expect to be the largest public offering on record, the ruling removes one specific and material risk, that a court-sanctioned blacklist could keep federal agencies and contractors away from its products, without settling the larger confrontation. Anthropic’s own disclosures to prospective shareholders are expected to treat both the pending appeals and the broader political mood surrounding artificial intelligence’s expansion into national security as material risks in their own right. It is a fitting coda to the case that a legal victory this complete still leaves the deeper argument, over how much authority a government may exercise over a contractor’s conscience, entirely unresolved.