A 2-1 ruling for the Department of War
The United States Court of Appeals for the District of Columbia Circuit denied Anthropic’s petitions for review on 25 September, upholding the Department of War’s decision to exclude Claude from its supply chain under the Federal Acquisition Supply Chain Security Act of 2018. The exclusion bars the department and its contractors from running Anthropic’s models on departmental information systems.
Circuit Judge Gregory Katsas wrote for the court, joined by Circuit Judge Neomi Rao. Circuit Judge Karen LeCraft Henderson dissented. The case, argued on 19 May, is numbered 26-1049 and was consolidated with 26-1162.
The department acted, the opinion says, “after Anthropic refused to relax contractual prohibitions on the use of Claude for lethal autonomous warfare or domestic surveillance”. Anthropic argued the exclusion was arbitrary, unauthorised by the statute and unconstitutional.
Why the majority sided with the department
“We reject these challenges,” Katsas wrote. “The Department had ample support for its conclusion that the continued integration of Claude into the Department’s information systems, by the Department or its contractors, presented a statutorily covered national-security risk.”
The reasoning turns on what Anthropic itself says about its product. “As Anthropic admits, the company encodes restrictions into Claude that prevent the model from performing tasks that Anthropic wishes to prevent,” the opinion says. “On more than one occasion, these restrictions have stopped Claude from performing tasks requested by government users.” The court also points to a recent dispute over whether Anthropic’s contract terms barred the use of Claude in an ongoing overseas military operation, which left the department “uncertain whether Claude would perform as needed and intended”.
The constitutional claims failed too. Due process was satisfied, the court held, because the department gave notice and a chance to respond. The First Amendment claim failed because the exclusion rested on Anthropic’s “refusal to assent to a contract term that the Department deemed essential, not based on the company’s support for greater governmental regulation of AI technology”.
The dissent reads the statute more narrowly
Henderson’s dissent is a question of statutory text. The Act lets the department act against a source that may “sabotage, maliciously introduce unwanted function, extract data, or otherwise manipulate” covered technology. Everything turns on that last phrase.
Anthropic read “manipulate” to mean subversive, deceptive acts. The department read it to mean operating or controlling something, as one manipulates a doorknob, “regardless of purpose or motive”, and the majority agreed. Henderson applied the canon that a word is known by the company it keeps: sabotage and malicious introduction both connote deliberate harm, so “manipulate” should too. “Because I believe that the context decidedly favors the narrower reading, I respectfully dissent,” she wrote.
Two courts, two answers
The ruling leaves the government with conflicting judgments. A federal judge in the Northern District of California held last month that a parallel designation was unlawful retaliation. The D.C. Circuit has now upheld the other one.
Anthropic said it respectfully disagrees with the decision and is weighing its options, including asking the full D.C. Circuit to rehear the case, Military Times reported. The company has said the designation has cost it billions in contracts, and it comes as Anthropic prepares to list.
The briefs supporting Anthropic were unusually broad: former defence secretary Leon Panetta, the American Civil Liberties Union and the Center for Democracy and Technology, 149 former judges, and employees of OpenAI and Google writing in their personal capacities.
The majority closed by naming the problem rather than solving it. The Secretary of War, Katsas wrote, “raises the deeply sobering prospect of overly constrained AI models shutting down unexpectedly and thus causing important military operations to fail”. Anthropic “raises the deeply sobering prospect of unconstrained AI models hallucinating inappropriate targets for lethal military force”. Both are national-security concerns, he wrote — but “in our Republic, it is the President and the Secretary of War who must determine how best to balance the competing risks.”
The next step is Anthropic’s: a petition for rehearing en banc, or the Supreme Court.
