Anthropic Pentagon Ruling: US Court Backs AI Supply-Chain Exclusion

A divided US appeals court has upheld the Pentagon's exclusion of Anthropic's Claude from its military supply chain. The ruling raises a wider question about control over military AI, while a separate California judgment means the legal picture remains narrower than a blanket ban.

By The Strategic Newb Editorial Team
Event: 25 September 2026 · Published: 26 September 2026

A US appeals court has upheld the Pentagon's decision to exclude Anthropic's Claude from its military supply chain after a dispute over safety restrictions. The 25 September ruling is consequential for defence AI contracts, but it does not settle a separate case involving a wider federal ban.

The Anthropic Pentagon ruling gives the US military a significant legal victory in its dispute with a leading AI developer. On 25 September 2026, a three-judge panel of the US Court of Appeals for the District of Columbia Circuit ruled two to one that the Pentagon could use a federal supply-chain security law to remove Claude from its own procurement chain. The judges rejected Anthropic's challenge to the designation, which had followed the company's refusal to remove restrictions on fully autonomous lethal weapons and mass domestic surveillance. Court opinion · Reuters

The ruling is a major setback for Anthropic, but its reach needs careful description. It concerns a supply-chain action under the Federal Acquisition Supply Chain Security Act and the Pentagon's own systems and contracts. Another federal court, in California, had already struck down a parallel designation under a different statute and blocked a government-wide ban. The Washington appeals court explicitly distinguished that case. It has not issued one definitive judgment governing all government use of Claude or all private use of the product. Court opinion · Breaking Defense

What did the Anthropic Pentagon ruling decide?

The central legal question was whether Anthropic's contractual and technical controls could count as a national-security “supply chain risk” for the military. The Pentagon argued that a model embedded in an operational system might fail to perform as expected if the supplier's restrictions blocked a requested use. Anthropic argued that its publicly stated safeguards were legitimate boundaries on uses it considered unsafe, and that the government was stretching a security statute designed to guard against subversion. Court opinion · Anthropic's February statement

The majority sided with the Pentagon. It concluded that the department had adequately supported its assessment that continued use could threaten military readiness and that less intrusive measures were not reasonably available. It also rejected Anthropic's due-process and First Amendment challenges. Judge Karen LeCraft Henderson dissented, arguing that the majority interpreted the law's definition of risk too broadly for a case involving an American supplier's declared safeguards. The vote was two to one, so the decision reflects an actual judicial disagreement over statutory boundaries, not a unanimous finding that AI guardrails are inherently dangerous. Court opinion · Reuters

Anthropic said it respectfully disagreed and was considering options, including further review. That is a company position, not a reversal of the ruling. The case could still be examined by the full appeals court or, if taken up, the Supreme Court. Neither has agreed to do so as of this edition. Breaking Defense · Reuters

Fact box

Decision: D.C. Circuit, Anthropic PBC v. United States Department of War, 25 September 2026.

Vote: Two judges for the Pentagon, one dissent.

Main effect: Pentagon supply-chain exclusion under a federal procurement statute upheld.

Separate case: A California judgment concerning a different designation and broader ban remains separate.

Next legal route: Anthropic may seek further review; no such outcome is guaranteed.

How a contract dispute became an AI governance test

Anthropic says it has supported US national-security work, including intelligence analysis, modelling, planning and cyber operations. Its two contested boundaries were narrower: it said it would not knowingly support fully autonomous weapons or mass domestic surveillance. The company argued that current frontier models are not reliable enough to select and engage targets without human control and that large-scale domestic surveillance raises civil-liberties concerns. These are Anthropic's safety and policy claims; the court was deciding the legality of the Pentagon's procurement decision, not scientifically proving a model's battlefield reliability. Anthropic

The Pentagon took a different view of who should set the terms of military use. Its position, as recorded in the opinion, was that the government needed assurance that a model it relies on would carry out all lawful authorised tasks without the supplier reserving an effective veto. The immediate clash came after contract talks failed in February and the department made a formal supply-chain determination in March. Anthropic challenged that determination in court. Friday's decision is the latest stage of that conflict, and the first point at which this appeals panel has resolved the legal challenge on the merits. Court opinion · Anthropic

The case is larger than one vendor because AI models are increasingly components within other software. A contractor might use a model to analyse information, draft plans or assist a workflow that serves a military customer. If the government regards supplier-set use restrictions as a risk to those workflows, vendors must decide how to write safety commitments that remain acceptable to defence buyers. Governments outside the United States will watch that balance: they too need predictable systems, accountable military decisions and credible limits on dangerous applications.

Who benefits, and who is under pressure?

The Pentagon gains a stronger legal basis to proceed with Claude's exclusion from its own supply chain. Other AI suppliers willing to accept the department's contract terms could see commercial opportunities, though the ruling does not award any contract to a rival. Defence contractors that integrated Claude into Pentagon work face possible transition costs and uncertainty about permissible use. The exact cost and timetable depend on individual contracts; neither should be guessed from this judgment alone.

Anthropic faces reputational and business pressure. Reuters reports that the company says the designation has cost it substantial business, but that amount is Anthropic's allegation, not a judicially established loss. Its ordinary enterprise customers should also avoid a broad misunderstanding: the military procurement ruling does not itself ban all private customers from using Claude, including firms that happen to hold defence contracts for unrelated work. Anthropic's earlier statement drew that same distinction. Reuters · Anthropic

Why the California case does not disappear

The two cases use different legal authorities. In August, a San Francisco federal judge rejected a separate Pentagon designation under a narrower statute and blocked elements of a broader federal exclusion. The D.C. Circuit's majority said that ruling did not control its review of the action under the procurement statute before it. Thus apparently conflicting headlines — one court striking down a designation and another upholding one — can both be accurate if they specify the statute and scope. Court opinion · Reuters

This distinction matters operationally. A buyer needs to know whether a restriction applies to a Pentagon contract, an unrelated federal contract or ordinary commercial use. The answer differs. Treating “Anthropic blacklisted” as a universal legal fact would misinform clients and readers. The more precise description is that one military supply-chain exclusion has survived appellate review while litigation over a wider policy follows a separate course.

What happens next?

Anthropic may ask the full D.C. Circuit to rehear the case or seek Supreme Court review. It may also continue contesting separate restrictions in California. The Pentagon must still manage any transition away from Claude in affected systems; the court did not design a replacement programme. The technology policy question remains unsettled even if this judgment stands: what contract terms let a government rely on AI for lawful missions while preventing uses the supplier and wider public consider unacceptable?

The Strategic Newb assessment

The answer to the reader's central question is specific. The 25 September decision permits the Pentagon's procurement exclusion of Claude under the law the court examined. It does not amount to a blanket US ban on Claude or a settled legal rule that all AI safety limits are incompatible with defence work. The deeper change is that a court accepted military readiness as a basis for treating supplier-controlled model restrictions as a supply-chain concern. Other AI companies now have a clearer warning about the contract terms they may face.

FAQ

Is Claude banned for everyone in the United States? No. The decision concerns the Pentagon's supply chain under a particular procurement law.

What safeguards caused the dispute? Anthropic's restrictions on fully autonomous lethal weapons and mass domestic surveillance.

Why did the court call this a supply-chain risk? The majority accepted the Pentagon's view that supplier controls could stop an integrated model from performing a task the military expects.

Can Anthropic appeal? It is considering further review, including a request to the full D.C. Circuit; acceptance and outcome are uncertain.

Sources

• US Court of Appeals for the D.C. Circuit, “Anthropic PBC v. United States Department of War, Nos. 26-1049 and 26-1162”, 25 September 2026 — primary decision, majority and dissent, precise legal scope.
• Reuters, “US appeals court upholds Pentagon's blacklisting of Anthropic”, 25 September 2026 — independent report of ruling, Anthropic reaction and California case.
• Breaking Defense, “DC Circuit panel upholds Pentagon's ban on Anthropic – so what comes next?”, 25 September 2026 — specialist independent context and statements from both sides.
• Anthropic, “Statement from Dario Amodei on our discussions with the Department of War”, 26 February 2026 — primary account of the company's two restrictions and its position on military use.
• Anthropic, “Where things stand with the Department of War”, 5 March 2026 — company account of designation and asserted scope.