Anthropic wins first court fight over Pentagon risk designation

Introduction

Anthropic has won its first substantive legal victory in its dispute with the U.S. Department of Defense. A federal court in California ruled that the Trump administration acted unlawfully when it designated the AI company a “supply-chain risk.” The decision does not resolve the entire conflict: Anthropic’s separate case in Washington, D.C., remains active. It does, however, impose an important legal constraint on how the government can restrict an AI company’s public-sector relationships in the name of national security.

Key points

• The risk designation was rejected. U.S. District Judge Rita Lin said Defense Secretary Pete Hegseth’s action amounted to “unlawful retaliation” in violation of the First Amendment. She also described the decision as arbitrary and capricious.
• The government failed to provide due process. The court found that Anthropic was not given the procedural protections required by the Fifth Amendment. The resulting order extended beyond the Defense Department and instructed all federal agencies to stop working with the maker of Claude.
• The dispute centered on model safeguards. Anthropic had drawn firm lines around the use of its systems for fully autonomous weapons and mass surveillance of American citizens. The Pentagon said it would use the models only for lawful purposes, while arguing that Anthropic might try to control military use after delivery.
• The government’s broader conduct appeared inconsistent. Lin pointed to the contrast between the supply-chain label and other government actions. The Defense Department continued pursuing a contract with Anthropic, worked with its new Mythos model on cybersecurity, and had even proposed applying the Defense Production Act to the company. That proposal suggested Anthropic was important to national security, not a threat to it.

Why the ruling matters

The case is about more than whether one AI company can sell its systems to the government. It tests whether a supply-chain security designation can be used as a policy lever against a company that refuses to accept certain applications of its technology. Lin wrote that an empty invocation of national security cannot serve as a blank check to punish or retaliate against government critics.

The ruling does not prevent the Pentagon from choosing another vendor. Lin explicitly recognized the department’s authority to select the AI supplier it prefers. But that procurement discretion does not authorize broad restrictions imposed without adequate evidence or lawful procedures. The judge also noted that Anthropic “undisputedly lacks” backdoor access once its technology is handed over to the Defense Department, weakening the government’s argument that the company could continue to control the models after purchase.

For the AI industry, the decision sends two signals. Safety commitments and limits on use may become central terms in government contracting, but they can also trigger political and procurement retaliation. At the same time, government action against AI suppliers must rest on evidence, consistent reasoning, and due process. Anthropic welcomed the ruling and said it remains focused on working with the government on national-security applications. The pending Washington case may determine how far the government’s authority extends.

Source: TechCrunch AI