A federal appeals court in Washington, D.C., ruled 2-to-1 on Friday that the Department of Defense has the legal authority to designate artificial intelligence developer Anthropic as a national security supply chain risk.
Key points
- A D.C. federal appeals court ruled 2-1 upholding the Pentagon's supply chain risk designation against Anthropic.
- The dispute stems from a collapsed $200 million contract after Anthropic demanded restrictions on autonomous weapons and domestic surveillance.
- Defense Secretary Pete Hegseth barred military agencies and defense contractors from deploying Claude models under the Supply Chain Security Act.
- The appeals panel stayed immediate enforcement to permit Anthropic time to request an en banc rehearing or Supreme Court review.
The ruling upholds an order issued by Defense Secretary Pete Hegseth under the Supply Chain Security Act. That order bars military agencies and federal defense contractors from integrating Claude models into their technical infrastructure.
Dispute Over Claude AI Military Boundaries
The legal clash originated with a $200 million contract signed in July 2025. That agreement aimed to integrate Claude models into GenAI.mil, the Pentagon’s centralized generative software environment.
Negotiations between the government and Anthropic broke down after the developer set firm operational boundaries. The company refused to permit deployment of its models in fully autonomous lethal weapon systems or for mass domestic surveillance targeting American citizens. It also insisted on retaining a technical kill switch to disable model access if operators violated those usage conditions.
The Department of Defense rejected those constraints, demanding full access for all lawful military operations. When the vendor refused to remove its internal safeguards, Hegseth declared the company a supply chain risk, effectively canceling contractor access to the Claude ecosystem.
The D.C. Circuit Majority and Dissent
Judge Gregory Katsas authored the majority opinion for the U.S. Court of Appeals for the D.C. Circuit, joined by Judge Neomi Rao. Katsas wrote that the military held legitimate concerns regarding vendor-imposed controls during armed conflict.
The majority noted that overly constrained commercial models could shut down unexpectedly in the middle of active operations or suffer from outside manipulation. Katsas concluded that final operational authority must remain with the President and the Defense Secretary rather than private software suppliers.
Judge Karen LeCraft Henderson dissented from the decision. Henderson stated that a technology vendor enforcing upfront ethical boundaries does not represent a national security or supply chain threat under federal law.
Conflicting Rulings and Supreme Court Path
The decision creates contrasting outcomes across the federal judiciary. Last month, U.S. District Judge Rita Lin in San Francisco struck down a separate, government-wide restriction against the developer. Lin ruled that federal officials unlawfully retaliated against Anthropic due to its public positions on safety governance.
Following the D.C. ruling, Hegseth praised the appellate panel’s decision in a statement on X. The San Francisco developer stated that it respectfully disagrees with the court and is evaluating its legal options.
The company maintains that the defense blacklisting has caused billions of dollars in commercial damages as it prepares for an initial public offering. Even as Anthropic expands its frontier research operations, federal procurement remains closed to its models under the appellate order.
The D.C. Circuit panel delayed immediate enforcement of the ruling. This stay gives the company time to file a petition for an en banc rehearing before the full appeals court or seek review from the U.S. Supreme Court.
