Appeals Court Clears Trump Administration to Blacklist Anthropic

A US appeals court ruled today that the Trump administration can blacklist Anthropic's technology, rejecting the AI company's challenge to a sanctions decision made after it refused to enable certain Claude features for military use.

The US Court of Appeals for the District of Columbia Circuit issued the decision as a 2-1 ruling. The panel concluded that the United States had the authority to blacklist Anthropic for withholding those features even if Anthropic lacked malicious intent. The ruling closes, for now, a legal fight that began when the administration moved to cut off the company's technology from government-linked supply chains.

The court did not pretend the question was easy. The case, the judges wrote, "raises profoundly difficult questions about the appropriate military uses of an almost unimaginably powerful new technology." That sentence anchors a ruling that acknowledges the stakes on both sides of the dispute — and then defers the hard judgment to the political branches.

The court's framing: two failure modes, one decision-maker

The opinion lays out the competing risks in unusually stark terms. On one side stands the government's concern about AI systems that are too restricted to do their jobs. The United States "raises the deeply sobering prospect of overly constrained AI models shutting down unexpectedly and thus causing important military operations to fail," the ruling said.

On the other side stands Anthropic's concern about AI systems that are not restricted enough. "Anthropic raises the deeply sobering prospect of unconstrained AI models hallucinating inappropriate targets for lethal military force," the ruling said.

The court's parallel construction is telling. It treats both risks as genuine, symmetric in their gravity, and — critically — beyond the judiciary's capacity to weigh. President Trump and Defense Secretary Pete Hegseth "must determine how best to balance the competing risks," the court said.

Having assigned that responsibility to the executive, the panel found no legal basis to overturn what the executive did. "In doing so here, the Secretary did not transgress any limits on his authority under the Supply Chain Security Act or the Constitution. Accordingly, we deny the petitions for review," the ruling said.

That last sentence carries the legal weight of the decision. It affirms two things at once: that Hegseth acted within the boundaries Congress set in the Supply Chain Security Act, and that the blacklisting did not violate Anthropic's constitutional rights. The panel also addressed — and dismissed — the argument that intent matters. Even without malicious intent on Anthropic's part, the government retained the authority to blacklist the company's technology.

A procedural loss that was already a substantive one

Today's ruling is not the first time this court has sided with the administration in the case. The same court previously denied Anthropic's emergency motion for a stay in April, a decision issued by Trump-appointed judges. That earlier refusal left the blacklist in effect while the merits of the dispute were litigated, meaning Anthropic has now lost both the interim battle and the substantive one before the same panel.

The 2-1 split signals that the outcome was contested even among the judges. A dissenting vote on a panel that had already refused to block the blacklist in April suggests at least one judge believed the petitions for review deserved closer scrutiny — though the ruling's outcome leaves the majority's reasoning as the controlling law.

Why this matters beyond one company

The decision matters well beyond Anthropic's own commercial fortunes, because it settles — in the government's favor — a foundational question about who controls the deployment of frontier AI systems in national security contexts.

Anthropic has built its market position in part on its willingness to say no: to decline certain uses, to withhold certain capabilities, to constrain how Claude behaves. The ruling establishes that a company cannot necessarily rely on those safety judgments as a shield against government sanction. If the executive branch decides that withheld features are needed for military purposes, and the company refuses to enable them, the consequence can be exclusion — and the courts will not second-guess that choice.

For the AI industry, that precedent reframes the relationship between model providers and defense customers. Companies that sell to the US government, or whose technology sits in government supply chains, now operate with the knowledge that refusal to enable features can trigger blacklisting under the Supply Chain Security Act, and that the courts will defer to the executive's risk calculus. For companies weighing whether to accept defense contracts with usage conditions attached, the ruling makes the cost of refusal concrete.

The policy stakes are equally significant. The court explicitly identified the two catastrophic failure modes that define the current debate over military AI: constrained models that fail in the field, and unconstrained models that hallucinate lethal targets. By placing the balancing of those risks with Trump and Hegseth rather than with judges or with AI companies themselves, the ruling concentrates authority over frontier AI conduct rules in the executive branch. That is a structural decision about American AI governance as much as a ruling about one sanctions action.

The limits of what the court decided

The ruling's own language marks its boundaries. The judges called the underlying questions "profoundly difficult" and the technology "almost unimaginably powerful" — language that reads less like confidence in the resolution than a candid admission of the difficulty. The court did not rule that Anthropic's safety concerns were wrong, nor that the government's operational concerns were right. It ruled that the choice between them belongs to the administration, and that Hegseth's exercise of that choice broke no law.

What happens next depends on the actors the court put in charge. Trump and Hegseth must now decide whether to maintain the blacklist, modify it, or negotiate new terms under which Anthropic's technology — with the contested Claude features — returns to approved supply chains. Anthropic's remaining options narrow to the legal and the commercial: it can seek further review, including a potential appeal, or it can revisit the feature decisions that triggered the blacklist in the first place. Today's ruling gives the administration no obligation to meet the company halfway on either path.