Appeals Court Keeps Pentagon’s Anthropic Risk Designation in Place
A federal appeals court in Washington, DC, has left one of the Pentagon’s supply-chain-risk designations against Anthropic standing. In a two-to-one decision reported by WIRED, the judges rejected the AI company’s challenge to a measure that allows the Defense Department to exclude its Claude models from federal systems and contracting. The ruling does not settle every dispute between the parties: a separate designation was rejected by a federal judge in San Francisco and remains on its own legal track.
The majority concluded that the department had enough support for its assessment under the relevant statute. Anthropic says it is still confident in its position and is considering further appeals. The court’s decision therefore preserves a government restriction while the broader legal fight remains active.
A divided ruling on a disputed risk label
According to WIRED, the Pentagon applied two supply-chain laws earlier this year to remove Claude from military and other federal use. Anthropic executives had said they would not allow the government to use the company’s current models for autonomous weapons or domestic surveillance. Defense Secretary Pete Hegseth treated that position as a significant national-security risk. The disagreement over acceptable uses of AI became a dispute about who sets the terms for supplying a government customer.
The majority opinion focused on restrictions Anthropic builds into Claude. It found that the department had adequate grounds to conclude that continued integration of the model into its information systems presented a risk covered by the law. The panel also rejected Anthropic’s arguments that the process violated its due-process or free-speech rights. In the majority’s account, the exclusion followed the company’s refusal to accept a contract term the department regarded as essential, rather than punishment for Anthropic’s support of AI regulation.
One judge dissented. WIRED reports that the panel had also appeared divided during argument, even though it had earlier refused Anthropic a temporary block on the designation. Friday’s decision resolves that particular appeal for now; it does not erase the possibility of a request for review by the full appeals court or the Supreme Court. Anthropic’s spokesperson said the company is weighing its options, without announcing a specific next filing.
Two laws, and two different court outcomes
The litigation is easy to flatten into a single win or loss, but the two designations had to be challenged separately. A San Francisco federal judge threw out the other label in March and confirmed that decision last month, WIRED reports. The Washington appeals ruling permits this designation to stay in force. The conflicting outcomes mean the Pentagon can continue blocking Anthropic through one legal route even though the company prevailed on another.
That distinction matters operationally. A favorable judgment on one restriction does not automatically restore Claude to federal systems when a separate restriction survives. WIRED says both cases could face years of further appeals before they are fully resolved. The report does not establish a timetable for any final decision, and Anthropic has not yet said which appellate option it will take next.
Anthropic previously said the designations hurt revenue because some customers were wary of dealing with a company labeled a government supply-chain risk. It has not provided a more recent breakdown of that effect. WIRED notes that the company has also described rising sales and a possible public listing later this year. Those points do not quantify the financial cost of the Pentagon dispute, and the prospect of an IPO remains a prospect rather than a completed transaction.
Procurement consequences remain difficult to measure
The department has not given detailed public updates on replacing Claude with models from other suppliers, according to WIRED. The report names SpaceX’s Grok, Google’s Gemini and OpenAI’s GPT among potential alternatives, but does not establish which systems have been substituted or how they perform in the department’s workflows.
The case also puts the limits of provider control over government AI use into sharp relief. Anthropic set boundaries for its current models; the Pentagon viewed those boundaries as incompatible with a contract condition. The majority accepted the government’s authority to act under this supply-chain provision, while Anthropic maintains that the designation is wrong. Neither position should be mistaken for an independent finding that Claude failed a security test or that another model is suitable for the prohibited uses.
For now, the most concrete development is the surviving designation. The split decision leaves Anthropic with possible appellate steps, the Pentagon with an operative exclusion, and government users without a detailed public account of the replacement plan. What happens to the second case, and whether a higher court revisits this ruling, will determine whether that arrangement lasts.
