A federal appeals court in Washington, DC on Friday upheld the Pentagon's designation of Anthropic as a supply chain risk, allowing the Department of Defense to keep Claude out of military and federal information systems. The judges split 2-1, finding the department had ample support to treat continued use of Claude as a covered national-security risk. With one of two sanctions now left in place indefinitely, the ruling sets a precedent for how AI usage policies affect government contracts.
How the two Pentagon sanctions split in court
Earlier this year the Pentagon acted under two separate supply-chain laws to remove Claude from the military and other federal systems by this month. The labels had to be challenged in separate courts. A federal judge in San Francisco threw out one designation in March and confirmed that ruling last month. Friday's DC Circuit decision keeps the second designation in force indefinitely, so the Pentagon block can continue. Spokesperson Danielle Cohen said the company is considering all options, including a broader DC Circuit panel or the Supreme Court.
The core of the Pentagon's case is that Anthropic encodes restrictions into Claude that block tasks the company wants to prevent. Anthropic executives said the company would not allow current models to support autonomous weapons or domestic surveillance. Secretary of Defense Pete Hegseth called that stance a significant national security risk. The majority said the exclusion rested on refusal to accept a contract term deemed essential, not on support for regulation, and rejected due process and free speech claims as standard contract negotiations that followed procedure.
In April the same panel had declined to temporarily block the designation, finding Anthropic had not met stringent requirements for immediate relief. At the hearing, the three judges pressed both sides and appeared divided on a full block. The Pentagon has given no detailed update on replacing Claude with SpaceX's Grok, Google's Gemini or OpenAI's GPT models. Some staff at Google and OpenAI objected on ethical grounds, but the companies set protests aside and called support for the government crucial.
What this means for AI vendors and buyers
For contractors and agencies that use Claude, the surviving designation means a forced vendor switch by this month. Large integrators can shift workloads to Grok, Gemini or GPT models, though migration still requires retesting and new security reviews. Smaller suppliers that built processes around Claude face higher switching costs and fewer engineers to redo integrations. Anthropic said it lost revenue after the designations as customers feared dealing with a government pariah, which shows how procurement risk spreads beyond federal contracts.
Much remains unclear for buyers. Anthropic has given no further update on the bottom-line impact while reporting growing sales, so the lasting commercial effect is unproven. The Pentagon has not detailed replacement progress or how alternatives handle the weapons and surveillance tasks at issue. With one label tossed in San Francisco, the other upheld in Washington and years of appeals possible, buyers should clarify how usage limits are encoded, which contract terms are non-negotiable, and what fallback applies if a model leaves approved lists.
The trend will be tested by two markers: whether Anthropic seeks review by the full DC Circuit or the Supreme Court, and whether it proceeds with a potential IPO later this year. Pentagon statements on replacing Claude and any update on sales impact would show whether the block reshapes the federal AI supplier list or remains a prolonged legal standoff.
