A ruling with consequences beyond one contract dispute
A federal appeals court on Friday upheld the Pentagon’s decision to designate Anthropic a national security supply-chain risk, handing the government a major victory in its dispute with the artificial intelligence company over how its Claude models may be used by the military.
The September 25 ruling by the U.S. Court of Appeals for the District of Columbia Circuit rejected Anthropic’s challenge to the Department of Defense’s exclusion of the company from its supply chain. The 2-1 decision said the department had adequate grounds under federal law to conclude that continued use of Claude in military information systems posed a covered risk. The decision leaves the Pentagon’s restriction in place while sharpening a consequential question for the AI industry: how far can a company’s limits on military use affect its eligibility for government work?
You may also like
Anthropic has said it will not permit Claude to be used for lethal autonomous warfare or domestic surveillance. The Pentagon argued that those contractual restrictions could prevent the model from performing tasks the military needs. The court’s majority agreed that the department could consider the restrictions and the uncertainty they created when assessing supply-chain risk.
Why the court sided with the Pentagon
In its opinion, the appeals court said Anthropic encodes restrictions into Claude that can stop the model from carrying out certain requests. The judges pointed to instances in which the safeguards had prevented government users from completing requested tasks and to a dispute over whether Anthropic’s contract terms applied to an ongoing overseas military operation. That dispute, the majority said, left the department uncertain whether Claude would perform as required.
The court rejected Anthropic’s argument that the government had exceeded its authority under the Federal Acquisition Supply Chain Security Act. It also dismissed the company’s constitutional claims. On due process, the majority concluded that Anthropic had received notice and an opportunity to contest the designation. On free speech, it said the exclusion rested on the company’s refusal to accept a contract term the department considered essential—not on Anthropic’s public position about AI regulation.
Judge Karen Henderson dissented. The split underscores that the ruling was not a unanimous endorsement of the government’s approach. But the majority’s decision provides the Pentagon with judicial backing for its use of supply-chain authority in this dispute and makes Anthropic’s immediate challenge in this court unsuccessful.
A conflict about control as much as capability
The case turns on a distinction that is becoming central to AI procurement: whether a model provider’s safety rules are simply product features, or whether those rules can create operational risks for a customer that needs predictable performance. Anthropic’s position is that restrictions on specified uses are deliberate guardrails. The Pentagon’s position, accepted by the majority, is that restrictions embedded in a system can matter to national security when they could interfere with military operations.
The decision does not establish that every AI company must accept every government use of its products. It addresses the Pentagon’s particular designation and the record before the D.C. Circuit. Nor does it resolve the broader policy debate over autonomous weapons or surveillance. But it offers a clear warning to companies pursuing defense business: contractual limits and technical safeguards may be evaluated not only as ethical commitments, but also as potential dependencies or vulnerabilities from the government’s perspective.
The dispute has also unfolded on more than one legal track. A separate federal case in California produced a ruling against the government’s actions, creating a split across proceedings. Friday’s D.C. Circuit decision concerns the Pentagon-related designation before that court; it should not be read as erasing the separate case or settling every dispute between the company and federal agencies.
Anthropic has argued that the blacklist has harmed its business and reputation. The precise financial consequences remain the company’s claim, not a finding made by Friday’s court. The decision nevertheless increases uncertainty for contractors and technology firms that rely on Claude or work with Anthropic, particularly where defense-linked systems are involved.
What to watch next
The immediate practical effect is that the Pentagon can maintain its exclusion of Anthropic from the department’s supply chain under the challenged action. The ruling does not decide how other agencies will handle Anthropic products in every circumstance, and it does not prevent the company from pursuing further legal options. The divided panel leaves open the possibility of additional appellate proceedings, though Friday’s opinion itself does not predict their outcome.
For the wider AI market, the takeaway is less about which model is strongest than about who controls the terms of deployment. As governments buy increasingly capable systems, providers are likely to face pressure to define—and negotiate—the boundaries around weapons, surveillance, and other sensitive uses. The court’s ruling gives the Pentagon leverage in that negotiation: in this case, it treated a supplier’s refusal to loosen those boundaries as a risk it could act on.
That makes the Anthropic case a test of the relationship between AI safety commitments and public-sector procurement. The decision does not settle where that boundary should ultimately lie. It does show that companies seeking military contracts may have to defend their safeguards not only as responsible design, but also against a government argument that those safeguards make a system less dependable for its intended mission.
Source: Reuters — www.reuters.com/world/us-appeals-court-declines-block-pentag...
Comments
0 comment