A US appeals court has allowed the Department of Defense to keep Anthropic on a supply-chain risk blacklist, ruling that the government may act when a supplier restricts how its technology can be used even without malicious intent. The 2-1 decision leaves federal agencies and defense contractors facing restrictions on business with the AI company.
The dispute began after Anthropic refused to enable some Claude capabilities for military use. The court said officials must balance two risks: a constrained model failing during an operation and an unconstrained model hallucinating an inappropriate target. It found that a broad federal procurement statute covers conduct that can deny or disrupt a product’s operation, not only sabotage by an adversary.
The ruling does not fully settle the case. A federal district judge previously found the designation illegal under a different law whose definition of supply-chain risk implies malicious action. Anthropic says it is considering further review, which could include asking the full appeals court or the Supreme Court to hear the dispute. For AI suppliers, the split shows how model-use restrictions can become procurement risks even when they are framed as safety controls.