- Published on
Court rules Trump can blacklist Anthropic for refusing to enable Claude features
- Authors
- Name
- DeepHorus
A U.S. appeals court has upheld the Defense Department’s decision to blacklist Anthropic, clearing the Trump administration to restrict federal use of the company’s AI technology because it would not provide certain capabilities for military operations. The 2-1 ruling matters because it gives the government broad latitude to manage dual-use AI risk—even when the vendor’s refusal reflects a safety policy rather than malicious intent.
The U.S. Court of Appeals for the District of Columbia Circuit said Anthropic’s challenge raised “profoundly difficult questions” over military uses of an almost unimaginably powerful technology. The panel weighed two competing dangers: overly constrained models could shut down unexpectedly and disrupt important missions, while unconstrained systems could hallucinate inappropriate targets for lethal force. It said President Trump and Defense Secretary Pete Hegseth must balance those risks.
Under the Supply Chain Security Act, the administration can act against technology viewed as a risk to U.S. defense supply chains. Anthropic sued in March after the government ordered federal agencies to stop using its products and barred defense contractors from doing business with the company. The appeals panel concluded that Hegseth had not exceeded his statutory or constitutional authority, denying Anthropic’s petitions for review; the court had also rejected an emergency stay request in April.
The dispute exposes a fundamental technical tension in frontier AI: model restrictions can make systems less capable, unpredictable or prone to refusing valid tasks, while looser controls can increase unsafe outputs. For military users, reliability, refusal behavior, target identification and safeguards around lethal decisions are not merely product features. They are operational concerns that can determine whether an AI-assisted mission completes, is delayed or fails.
Two of the three judges sided with the government: Gregory Katsas, a Trump appointee who previously served as deputy counsel to the president, and Neomi Rao, who worked in the Trump administration’s Office of Management and Budget. The majority said the secretary, not the courts, should decide how to weigh model availability against misuse. That allocation of authority is significant for every AI supplier seeking to set limits on defense deployments.
Anthropic said it respectfully disagreed and noted that another federal court had already found the government’s parallel designation unlawful. The company said it remained confident in its position and was considering further review. Its possible next steps include asking the full appeals court to reconsider the case or petitioning the Supreme Court. The market and developer community may also watch closely because the ruling could influence how model providers negotiate access, safeguards and liability in government contracts.
Despite the litigation, Commerce Secretary Howard Lutnick recently said the administration and Anthropic had patched up their relationship and were “in tune.” Practically, the decision may accelerate procurement disputes as contractors and agencies determine which models can be used under defense rules. In the longer term, the case could become a test of whether emergency national-security powers permit the government to pressure private laboratories into enabling capabilities they consider unsafe.
