Part of the AI Regulation News hub.
A divided D.C. Circuit panel has denied Anthropic's petitions for review, leaving in place the Department of War's section 4713 determination excluding Claude from that Department's supply chain
This is a panel decision in one dispute, reviewed on a deferential standard, and the dissent disagreed with the majority's reading of the statute. The question that travels to other contracts is narrower than the headline: whether a supplier's enforcement of its own usage restrictions can count as a "supply chain risk" under 41 U.S.C. 4713. The majority said it can, at least as applied here.
Bottom line: A court decision, not a rule. A three-judge panel of the D.C. Circuit, over a dissent by Judge Henderson, denied Anthropic's petitions for review on 25 September 2026, so the Department of War's March 2026 exclusion under 41 U.S.C. 4713 stands. It binds the parties to this case; we have not checked whether any rehearing petition has been filed.
Who this affects: Federal contracting officers who may be asked to implement a section 4713 determination, government-contracts counsel advising suppliers or primes, and compliance and security leads at federal contractors that build commercial AI models or usage policies into work for an agency.
Issue date: Decided 25 September 2026; argued 19 May 2026. The Secretary's determination was made 3 March 2026, and the Department's 6 March memorandum set an outer limit of 180 days for removal. The opinion sets no new deadline.
What changed: The court rejected Anthropic's statutory and constitutional challenges. It held that the Secretary's concern about Anthropic disabling Claude from performing lawful actions requested by the Department qualifies as a "supply chain risk" under section 4713, and that no bad motive is needed under that definition. The court also recorded that once a model is delivered for use on the Department's classified systems, Anthropic cannot "access it, alter it, or shut it down", and that the Secretary's determination "did not depend" on post-deployment access or control; the concern it upheld was restrictions Anthropic can encode in each new version it delivers.
Analysis: On our reading, the holding rests on the majority's plain-meaning reading of "manipulate" in 41 U.S.C. 4713(k)(6), which it said turns on what a supplier does, not why. The court distinguished 10 U.S.C. 3252, whose narrower definition it read as needing bad motive. Which statute an agency invokes therefore matters to how a challenge is decided.
Primary sources: Anthropic PBC v. U.S. Department of War, No. 26-1049, opinion of 25 September 2026 (D.C. Circuit, PDF)
- Instrument (EN)
- Anthropic PBC v. United States Department of War and Peter B. Hegseth, in his official capacity as Secretary of War; opinion for the court and dissenting opinion
- Authority
- United States Court of Appeals for the District of Columbia Circuit (Henderson, Katsas and Rao, Circuit Judges; opinion for the court by Judge Katsas; dissent by Judge Henderson)
- Jurisdiction
- United States, federal. Petitions for review of agency action of the Department of War, under the review provision in 41 U.S.C. 1327(b)
- Status
- Decided 25 September 2026. The opinion ends: "Accordingly, we deny the petitions for review."
- Bindingness
- A court judgment binding the parties. It reviews one Department's determination about one supplier; it is not a rule and does not address other agencies' procurement
- Issue date / next deadline
- Decided 25 September 2026. No new deadline set by the opinion. The removal memorandum of 6 March 2026 said "as soon as practical," and in any event within 180 days
- Statute
- Federal Acquisition Supply Chain Security Act of 2018: 41 U.S.C. 4713 (covered procurement actions) and 41 U.S.C. 1327 (judicial review in the D.C. Circuit)
- Case numbers
- No. 26-1049, consolidated with No. 26-1162 (the second petition, filed 17 June 2026, consolidated 24 June 2026)
- Primary source
- https://media.cadc.uscourts.gov/opinions/docs/2026/09/26-1049-2194984.pdf
What did the D.C. Circuit decide?
It denied both petitions. This is a panel decision of the D.C. Circuit in a single case, and it leaves the Department of War's exclusion in place. Writing for the court, Judge Katsas framed the dispute in the opinion's opening paragraph: the Department decided to exclude Claude "from its supply chain under the Federal Acquisition Supply Chain Security Act of 2018" after Anthropic "refused to relax contractual prohibitions on the use of Claude for lethal autonomous warfare or domestic surveillance."
The holding, in the court's words, follows directly: "We reject these challenges. 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." That is the court's conclusion about the record before the Secretary, reviewed on the standard described below.
The court also rejected the constitutional claims. On due process, it said the Department "promptly notified the company of the exclusion and its supporting rationale, and then gave the company a fair opportunity to contest the exclusion." On the First Amendment, it held that the record showed the Department acted "not because of its advocacy, but because Anthropic refused to agree to a contract term the Department deemed essential to national security."
Judge Henderson dissented on the statutory question. Her opinion is summarised in its own section below.
What does section 4713 allow, and how did the Department use it here?
As the court describes it, 41 U.S.C. 4713 authorizes "covered procurement action[s]" to prevent agencies from using information technologies that pose a risk to national security. Those actions, the opinion says, include barring agency contracts with a particular supplier and subcontracts that use the supplier to perform work for the agency.
Before acting, the agency head makes a written determination that using the authority "is necessary to protect national security by reducing supply chain risk" and that "less intrusive measures are not reasonably available". Notice and an opportunity to respond normally come first, but the statute lets the agency head delay notice on an "urgent national security interest". Review runs to the D.C. Circuit on a petition filed within 60 days of notification, and the court reads 41 U.S.C. 1327(a) as barring other judicial review of action taken under section 4713.
The dates, all taken from the opinion: on 3 March 2026 the Secretary determined that use of Claude in Department systems "presents a significant supply chain risk", that no less intrusive measures were reasonably available, and that an urgent national security interest required immediate action. Notice was dated 3 March and emailed on 4 March. On 6 March the Department's Chief Information Officer ordered removal of Anthropic products "as soon as practical," and in any event within 180 days, and the court says that memo "further prohibited contractors from using Anthropic products in their work for the Department." Anthropic petitioned on 9 March. A supplemental notice followed on 19 March, the court denied a stay but expedited the case on 8 April, and the Secretary denied reconsideration on 3 June.
The court read the statute as allowing the determination and the covered procurement actions to happen at the same time, which is how it found its own jurisdiction secure from the 9 March petition.
How did the majority read "supply chain risk"?
Section 4713(k)(6) defines supply chain risk to include the risk that any person may "otherwise manipulate the design, integrity, manufacturing, production, distribution, installation, operation, maintenance, disposition, or retirement" of covered articles "so as to surveil, deny, disrupt, or otherwise manipulate the function, use, or operation" of them. Anthropic argued that the definition carries a requirement of hostility or bad motive. The majority disagreed.
It gave "manipulate" its ordinary meaning and concluded: "at least as applied here, the statutory definition of a "supply chain risk" turns on what Anthropic does, not why Anthropic does it." It added that section 4713 imposes no criminal or civil liability, describing it as "a procurement statute addressed to what goods and services the Department may buy".
The factual footing, as the court tells it, was that Anthropic's model training "does effectively enforce contractual usage restrictions", together with a dispute over whether those restrictions reached a military operation abroad. The opinion also states a limit. Anthropic told the court that once a model is delivered for use on the Department's classified systems, it cannot "access it, alter it, or shut it down", and the Secretary clarified that his determination "did not depend" on Anthropic's "real-time technical access to or control" of Claude after deployment. The majority upheld the exclusion anyway, because, in its words, "Anthropic can and does program Claude's behavior with each new model it delivers to Department contractors". The court was explicit about what it was not deciding: "the point here is not that these model or usage restrictions were arbitrary". Its conclusion was that "the Department reasonably feared that Anthropic might manipulate Claude's design to prevent it from performing national-security functions that the Department deems contractually authorized and necessary."
The court contrasted 10 U.S.C. 3252. It noted that a district court decision, Anthropic PBC v. U.S. Dep't of War, No. 26-cv-01996 (N.D. Cal. Aug. 27, 2026), had set aside a separate designation under section 3252, and said it had "no quarrel" with that court's reading that bad motive is required under section 3252's narrower definition. "But as explained at length above, no such bad motive is required to support a designation under the much broader definition set forth in section 4713." A footnote adds that the Northern District's judgment does not control review of the section 4713 action.
What standard of review did the court apply?
The opinion states the standard in its own terms, and we repeat it rather than characterise it. Under 41 U.S.C. 1327(b)(2), the court holds unlawful covered procurement actions it finds "arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law." On arbitrariness, it said it requires only that agency action be "reasonable and reasonably explained", and that review is "deferential".
Purely legal questions of statutory construction it reviews de novo, but "In construing statutes implicating national security, we resolve ambiguities against intruding on the national-security determinations of the Executive Branch." And "where a factual question turns on assessments of national security, we give the agency more deference, even if constitutional claims are at issue."
That standard shaped the two remaining statutory points. On less intrusive measures: "The Secretary reasonably concluded that less intrusive measures were not reasonably available, and we cannot override that fact-based, national-security assessment." On urgency, the court did not decide whether the Secretary's finding was right. It assumed for argument that the finding was reviewable and wrong, applied a harmless-error rule, and held that "Anthropic has suffered no prejudice" because the post-exclusion process had already run and the Secretary had maintained the exclusion after considering Anthropic's full submission.
What did Judge Henderson's dissent say?
The dissent turns on one phrase. Judge Henderson wrote that whether the Secretary lawfully invoked the statute turns on whether Anthropic falls within its definition of supply chain risk, and that the answer depends on the residual clause "or otherwise manipulate." She concluded: "Because I believe that the context decidedly favors the narrower reading, I respectfully dissent."
Her reading draws on the surrounding words. Applying noscitur a sociis, ejusdem generis and the series-qualifier canon, she read "maliciously" as reaching "extract data" as well as "introduce unwanted function", and read the verbs around "manipulate" as connoting intentionally hostile or clandestine acts. In her view the majority's reading leaves much of the definition's "motive-infused language" as surplusage.
She then described what she saw as the consequence of the majority's reading for suppliers generally: "It matters not if the restrictions are ones the Department has already agreed to. And it is immaterial if the contractor enforces the restrictions based on a good faith and legally correct interpretation of them." Of a future supplier asked to change its AI-use policies, she wrote: "According to today's decision, that contractor will have a choice: Agree to the Secretary's demands or risk being designated a national security threat under FASCSA."
Relying on the 2018 Senate report and pre-enactment agency material, she wrote: "I cannot agree that this is the scenario the Congress had in mind when it enacted FASCSA." That history, in her words, "refutes the view that "manipulat[ion]" of a covered article encompasses anything like the conduct that, under today's holding, gives rise to a supply chain risk", which she described as "a contractor's honest and upfront enforcement of restrictions on a covered article's use disfavored by the government." The majority, for its part, said the dissent's arguments "have some force, but we are ultimately unpersuaded", and declined to use legislative history to fix the ordinary meaning of the definition's verbs.
What does this mean for contracting officers and contractors?
What follows is our interpretation, not the court's. The opinion decides one Department's determination about one supplier, on a deferential standard, and it is a panel decision that could still be the subject of further review we have not checked. It does not exclude Claude from other agencies' procurement and does not purport to.
With that limit stated, three points from the text transfer. First, the majority treated a supplier's ability to enforce usage restrictions through the product itself as within section 4713's definition, without any need to show bad motive, "at least as applied here". Second, the court weighed the agency's choice of section 4713 over conventional contract-termination authorities and did not disturb it, observing that Anthropic "has not explained why it would suffer any less of a stigma" under those other authorities. Third, the court accepted that a section 4713 action can reach subcontracts, and it reports that the Department's implementing memorandum prohibited contractors from using Anthropic products in their work for the Department.
For contractors, on our reading, the practical question is how a model provider's usage policy, model-level restrictions and update cadence map onto the work a customer agency expects to do, and who resolves a dispute about scope. The court called the contested restrictions "hardly self-defining". The opinion does not tell contracting officers how to draft any clause, and we do not either.
What we did not verify
What we opened: the full text of the D.C. Circuit opinion in No. 26-1049 as extracted from the court's PDF (media.cadc.uscourts.gov), including the opinion for the court by Judge Katsas and the dissent by Judge Henderson. Every date, number and quotation on this page comes from that text. Straight quotation marks and apostrophes replace the court's typographic ones; no words were changed.
What we did not open: the appendix and record materials the opinion cites, the Secretary's 3 March 2026 determination, the 6 March 2026 memorandum, the 3 June 2026 reconsideration decision, the briefs and amicus briefs, the April 8 stay order, and the Northern District of California decision in No. 26-cv-01996. Where we describe those documents, we describe them only as the opinion does. We did not check the docket for any petition for rehearing or rehearing en banc, any mandate, or any petition for certiorari, and we did not confirm whether the opinion is designated for publication.
What we refuse to claim: we do not say the decision bars Claude across the federal government; it concerns one Department's determination. We do not say the court found Claude unsafe, defective or unreliable, or that it ruled on the merits of Anthropic's usage restrictions; the court said its point was "not that these model or usage restrictions were arbitrary". Every national-security conclusion on this page is attributed to the Secretary or to the court reviewing his determination under the statute. We do not characterise Anthropic, Claude or the Department beyond the opinion's own words, and we make no claim about how often section 4713 has been used.
One drafting point in the source: the dissent's text cites the definition as 10 U.S.C. 4713(k)(6) and in places as sections 4173 and 4716, while the majority cites 41 U.S.C. 4713(k)(6). We follow the majority's citation and have not checked whether a corrected version has been issued.
Disclosure: this page was drafted with the assistance of an AI model developed by Anthropic, a party to this case. For that reason it quotes the court's holding and the dissent verbatim, attributes every conclusion, and adds no assessment of the merits.
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
In this case, the D.C. Circuit read 41 U.S.C. 4713's definition of supply chain risk to reach a supplier's enforcement of its own usage restrictions through the product, regardless of motive, and reviewed the Department's national-security judgments deferentially. A dissent read the same words to require hostile or deceptive conduct. Anyone advising on AI usage policies in federal work would want to know which statute an agency is invoking, because the court treated 41 U.S.C. 4713 and 10 U.S.C. 3252 differently.
Source File
https://media.cadc.uscourts.gov/opinions/docs/2026/09/26-1049-2194984.pdf
Open the D.C. Circuit opinion in No. 26-1049 and confirm four things: the decision date of 25 September 2026 on the caption page, the holding in the second paragraph of Judge Katsas's opinion, the statement "Accordingly, we deny the petitions for review" at the end of Part VI, and the opening and closing of Judge Henderson's dissent.
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. ยท Anthropic PBC v. U.S. Department of War, D.C. Cir. No. 26-1049, opinion for the court by Judge Katsas, decided 25 September 2026
FAQ
Does this decision bar Claude across the federal government?
No. The opinion reviews the Department of War's exclusion of Claude from that Department's supply chain, made by its Secretary under 41 U.S.C. 4713. As the court describes it, the Department's 6 March 2026 memorandum ordered removal from its own systems and prohibited contractors from using Anthropic products in their work for the Department. The opinion does not address other agencies.
Did the court find that Claude is unsafe?
No. The court reviewed whether the Secretary's determination was lawful and reasonably explained. It said "the point here is not that these model or usage restrictions were arbitrary"; its finding was that Anthropic's model training enforces contractual usage restrictions, which the majority read as within the statutory definition of supply chain risk.
What standard of review did the court use?
In the opinion's words, review for arbitrariness is "deferential" and requires only that agency action be "reasonable and reasonably explained." Purely legal questions are reviewed de novo, but the court said it resolves ambiguities in national-security statutes against intruding on Executive Branch determinations, and gives "more deference" where a factual question turns on national security.
Is this the final word?
It is a panel decision denying the petitions for review, over a dissent by Judge Henderson. The opinion does not mention further review, and we have not checked the docket for any petition for rehearing, rehearing en banc or certiorari.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.