Judge Voids Pentagon’s Anthropic Ban as Retaliation

A federal judge has struck down the Pentagon’s designation of Anthropic as a national-security supply-chain risk, calling it unlawful retaliation for the company’s public criticism of the administration’s AI policy. U.S. District Judge Rita Lin, of the Northern District of California, issued the 59-page decision on August 28 and ordered the Defense Department to vacate the designation and rescind the directives built on top of it.

The designation dates to February 2026, when the Pentagon applied a procurement statute normally reserved for guarding federal systems against foreign sabotage. It followed Anthropic’s refusal to accept contract terms that would have required dropping two guardrails the company treats as non-negotiable: no use in fully autonomous weapons operating without human involvement, and no support for mass domestic surveillance. CEO Dario Amodei said at the time that Anthropic could not “in good conscience accede” to unrestricted military deployment of Claude. Anthropic sued in March.

Lin’s reasoning is unusually blunt about motive. The Pentagon’s actions, she wrote, “were based on a desire to make a public example out of Anthropic for its ‘arrogance’ in criticizing the government, not based on any articulable basis to believe that Anthropic would actually sabotage its model.” She described the government’s evidentiary submissions as slim, and noted they sat awkwardly beside Defense Secretary Pete Hegseth’s earlier invocation of the Defense Production Act — an act that treated Anthropic as essential to national security rather than a threat to it. The order finds both a First Amendment violation and a failure to provide the procedural protections the company was due, and holds the designation arbitrary and capricious.

The label had real commercial consequences. After Anthropic held its line, the Pentagon signed AI agreements with eight large vendors while cutting Anthropic out, and Google picked up sweeping classified access in the gap. A presidential social-media directive telling federal agencies to stop using Claude followed the designation; Lin had already blocked its enforcement on a temporary basis before this week’s ruling.

The fight is not over. A separate Anthropic case, turning on a different source of regulatory authority, is pending before a federal appellate court in Washington, D.C., and the government is widely expected to challenge Lin’s decision. Nothing in the order requires the Pentagon to buy Anthropic’s models; it removes the penalty, not the discretion. What it does remove is the mechanism that made Anthropic radioactive to any agency that wanted to keep working with it — a distinction that matters as the company prepares a public offering in which government exposure is a line item investors will read closely.

For a sector where “we have safety policies” is usually a marketing claim with no cost attached, the ruling attaches a price to the other side of the ledger. Anthropic spent six months blacklisted for enforcing two written rules, and a court has now said the enforcement was the protected part.