TL;DR: Appeals court keeps Pentagon's Anthropic risk tag in place

TL;DR: On Sept. 25, 2026, the U.S. Court of Appeals for the District of Columbia Circuit voted 2-1 to deny Anthropic's challenge and leave the Pentagon's supply-chain risk designation in place.

The ruling preserves Defense Department limits tied to Anthropic's Claude artificial-intelligence system while a conflicting California federal court decision remains in effect, according to ABC News and Ars Technica.

What the D.C. Circuit decided on Sept. 25

The court issued a divided decision Friday that rejected Anthropic's request to overturn the Defense Department's exclusion, according to ABC News.

The majority concluded that department officials had sufficient evidence to view Claude's embedded restrictions and an unresolved contract disagreement as a potential reliability problem for military operations, according to ABC News.

The judges also rejected arguments that the designation infringed free-speech protections or denied due process, according to ABC News.

The panel denied the petitions for review, leaving the procurement-related designation undisturbed under the authority the Pentagon invoked, according to Ars Technica.

The same appeals court had previously turned down Anthropic's emergency request for a stay in April, according to Ars Technica.

Why the Pentagon acted against Claude

Anthropic develops Claude, a family of large-language AI models that respond to prompts and can carry out tasks with limited ongoing direction, according to court materials summarized by Ars Technica.

The dispute followed Anthropic's refusal to loosen contractual limits on using Claude for lethal autonomous warfare or domestic surveillance, according to ABC News.

Pentagon leaders argued that those written limits and technical controls could prevent the system from completing lawful tasks sought by defense users, according to ABC News.

The majority said officials reasonably worried that design choices could keep Claude from performing national-security functions the department viewed as contractually authorized and necessary, according to Ars Technica.

Defense Secretary Pete Hegseth welcomed the decision in a post on X on Friday, writing, "Confirmed: @AnthropicAI = Supply Chain Risk," according to ABC News.

Hegseth added that the department then called the Department of War does what is right for the country and service members, according to ABC News.

Department spokesman Sean Parnell wrote separately that the department would never compromise on national security and that the ruling fully validated its position, according to ABC News.

Anthropic's retaliation and constitutional arguments

Anthropic argued that federal officials retaliated because the company opposed use of its technology for autonomous lethal operations or broad surveillance of Americans, according to ABC News.

The company said officials violated its First Amendment rights, misapplied a national-security procurement law to blacklist it, and bypassed ordinary procedures for ending federal contracts, according to ABC News.

The D.C. Circuit majority disagreed, finding that the department acted because Anthropic would not accept a contract term considered essential, not because the company favored stronger regulation of AI, according to court reasoning described by Ars Technica.

On due process, the court found no violation because officials gave prompt notice of the exclusion and supporting rationale and offered a fair chance to contest it, according to that same reasoning.

Anthropic said Friday that it disagreed with the appellate outcome and remained confident in its legal position, according to ABC News.

"Another federal court has already held the government's parallel designation unlawful. We remain confident in our position and are considering all options, including further review," a company spokesperson said, according to ABC News.

A 2-1 split and the dissent

Circuit Judge Gregory Katsas wrote the opinion for the court, with Circuit Judge Neomi Rao joining the majority, according to Ars Technica.

Both majority judges were appointed by President Donald Trump and served in the first Trump administration, with Katsas previously serving as deputy counsel to the president and Rao serving in the Office of Management and Budget, according to Ars Technica.

Judge Karen LeCraft Henderson dissented from the majority's interpretation of the procurement law, according to ABC News.

In dissent, she wrote that the statute does not treat "a contractor's honest and upfront enforcement of restrictions" as the type of supply-chain risk that allows blacklisting, according to ABC News.

Henderson was appointed by President George H.W. Bush, according to Ars Technica.

The majority opinion acknowledged the broader stakes, saying the case raised difficult questions about military use of powerful new technology, according to Ars Technica.

It contrasted the possibility of tightly constrained models stopping unexpectedly and causing operations to fail with the possibility of unconstrained models producing incorrect targets for lethal force, according to Ars Technica.

The court said Trump and Hegseth must weigh those competing dangers, and it found that the secretary stayed within the limits of the Supply Chain Security Act and the Constitution, according to Ars Technica.

Two courts, two different legal conclusions

Anthropic sued the administration in March after officials directed federal agencies to stop using its products and barred defense contractors from conducting business with the company, according to Ars Technica.

That litigation produced parallel rulings under different statutes, according to Ars Technica.

In August, U.S. District Judge Rita Lin in the Northern District of California ruled against the government under 10 U.S.C. Section 3252, according to ABC News and Ars Technica.

Lin said the government's actions "were based on a desire to make a public example out of Anthropic for its 'arrogance' in criticizing the government, not based on any articulable basis to believe that Anthropic would actually sabotage its model," according to ABC News.

That district-court decision remains in effect, according to ABC News.

The D.C. Circuit case examined the designation under a different law, 41 U.S.C. Section 4713, according to Ars Technica.

Section 3252 focuses on risks involving an adversary that may sabotage, maliciously introduce unwanted functions, or otherwise subvert a covered system, according to Ars Technica.

The appeals majority said it had no disagreement with the California court's conclusion that Anthropic acted without such an improper motive for purposes of Section 3252, according to Ars Technica.

But it held that Section 4713 uses broader language and does not require proof of a hostile motive, according to Ars Technica.

Why Section 4713 proved decisive

Section 4713 defines supply-chain risk to include the risk that any person may sabotage, maliciously introduce unwanted functions, extract data, or otherwise manipulate the design, integrity, manufacturing, production, distribution, installation, operation, maintenance, disposition, or retirement of covered technology products, according to Ars Technica.

The definition further covers manipulation affecting the function, use, or operation of those products or information stored or transmitted on them, including action that would surveil, deny, disrupt, or otherwise manipulate that function, according to Ars Technica.

The majority emphasized the phrase "any person," reading it as broader than adversaries or foreign entities, according to Ars Technica.

It also emphasized the word "deny," reasoning that withholding requested Claude capabilities could fall within that term, according to Ars Technica.

"In sum, we conclude that the Secretary's concern about Anthropic disabling Claude from performing lawful actions requested by the Department qualifies as a 'supply chain risk' within the meaning of section 4713," the majority wrote, according to Ars Technica.

Congress gave the D.C. Circuit exclusive jurisdiction over procurement actions taken under Section 4713 designations, according to Ars Technica.

What it means for defense technology procurement

For Pentagon buyers and contractors, the immediate effect is continued uncertainty because two federal rulings point in different directions under separate statutory authorities, according to ABC News and Ars Technica.

Anthropic told CNBC that it could seek en banc reconsideration by all active D.C. Circuit judges or ask the Supreme Court to review the decision, according to Ars Technica.

Commerce Secretary Howard Lutnick recently said the administration and Anthropic had repaired their relationship and were "in tune," despite the continuing court fight, according to Ars Technica.

For companies that supply software, data tools, and information systems for defense and national-security missions, the majority's reading preserves a broad Pentagon ability to act when built-in model limits could affect operational reliability, according to Ars Technica.

The dissent warns that treating openly disclosed use restrictions as a supply-chain threat could expand exclusion authority beyond sabotage-like conduct, according to ABC News.

The next procedural step belongs to Anthropic: pursue further appellate review or attempt to resolve the contract dispute that the majority identified as a source of operational uncertainty, according to Ars Technica.

Sources referenced