A divided federal appeals court has upheld a Pentagon decision to exclude Anthropic's Claude models from parts of its supply chain, handing the Trump administration a victory in one of two legal fights over the AI company's military-use restrictions.
In a 2–1 opinion issued September 25, the U.S. Court of Appeals for the D.C. Circuit denied Anthropic's petitions to overturn the action. The majority said the Department had enough support to conclude that Claude's built-in safeguards could stop the model from carrying out functions the military considered lawful and necessary.
The dispute grew out of Anthropic's refusal to relax contractual prohibitions on using Claude for lethal autonomous warfare or domestic mass surveillance. Anthropic told the court that it cannot access or alter models once they are deployed on classified Pentagon systems. The majority nevertheless focused on restrictions encoded during model training and on future model versions delivered through contractors.
Why the court sided with the Pentagon
The Federal Acquisition Supply Chain Security Act lets an agency bar a supplier from covered contracts and subcontracts when the agency determines that doing so is necessary to reduce a national-security risk and that less intrusive measures are not reasonably available.
Writing for the majority, Judge Gregory Katsas said the statute's reference to manipulating a product's “design” or “operation” can include Anthropic's openly imposed model limits; it does not require a malicious motive. Judge Neomi Rao joined the opinion. The panel also rejected Anthropic's due-process and First Amendment claims, concluding that the company received a fair chance to challenge the action and that the Pentagon acted over a contract term it considered essential rather than Anthropic's public advocacy for tighter AI rules.
Judge Karen LeCraft Henderson dissented. She argued that “manipulate” must be read alongside statutory terms such as sabotage and maliciously introducing unwanted functions, which in her view target deceptive or hostile conduct. Under that narrower reading, Anthropic's disclosed safety restrictions would not qualify as a supply-chain risk.
A separate California ruling still matters
The decision does not simply reverse the August ruling in Anthropic's favor. That case challenged a broader Pentagon designation and government-wide restrictions under a different legal authority. The D.C. Circuit majority said the California court's findings do not control this narrower procurement case, over which Congress gave the appeals court exclusive jurisdiction.
The practical result is a split legal landscape: one Pentagon measure was set aside in California, while the section 4713 supply-chain exclusion reviewed in Washington remains in force. The appeals-court opinion does not provide a detailed public accounting of how far the Pentagon has progressed in replacing Claude across its systems and contractors.
Technology