Court Backs Pentagon Ban on Anthropic Over Claude Safety Guardrails

A federal appeals court in Washington ruled on Friday that the Pentagon can keep Anthropic designated a supply chain risk, a 2–1 decision that leaves Claude barred from the Department of Defense. The label also stops any contractor, supplier or partner that does business with the U.S. military from doing business with the company. 12

The fight started on March 3, 2026, when Defense Secretary Pete Hegseth formally designated the company after months of failed negotiations. Two days later the department's chief information officer told its components to remove Anthropic's products from all of its systems and networks within 180 days. 23

Anthropic would not let Claude be used for lethal autonomous weapons or mass surveillance of Americans. It says AI is not yet reliable enough for autonomous weapons, and it calls domestic surveillance a violation of fundamental rights. The department wanted a contract term permitting "all lawful uses" instead, and when the company declined, it was designated. 13

The majority said the department "had ample support" for concluding that integrating Claude into its systems presented a national security risk. Anthropic, the court wrote, "encodes restrictions into Claude" that had already stopped the model from doing tasks government users asked for. 4 The record it leaned on includes a 2025 incident where Claude refused prompts from the Centers for Disease Control and Prevention, and a dispute over whether the contract barred its use in an ongoing overseas military operation. 23

The court then set out two dangers: an overly constrained model shutting down during an important military operation, and an unconstrained one hallucinating targets for lethal force. Weighing them, the ruling says, is the job of the president and the secretary of war. It rejected Anthropic's due process and free speech claims. Judge Karen Henderson dissented, arguing that the supply chain law was written to stop hostile foreign powers, not a US company enforcing its own guardrails. 24

Two laws were used, and two courts have now split on them. In August a federal judge in San Francisco struck down one of the designations as unlawful retaliation; Friday's ruling upheld the other, written under a broader statute that only this appeals court reviews. The Pentagon's ban therefore stands while the California case continues. 24

Anthropic says the designation has cost it billions of dollars in lost business and damaged its reputation ahead of an expected IPO. It is weighing a review by the full appeals court or a petition to the Supreme Court, both of which are discretionary. The Pentagon has been substituting other companies' models, including Google's Gemini and OpenAI's GPT. 14