Two Votes for the Pentagon: D.C. Circuit Upholds the Anthropic Blacklist

A three-judge panel of the U.S. Court of Appeals for the D.C. Circuit ruled 2-1 on Friday, September 25, 2026, that the Department of Defense may continue to treat Anthropic as a supply-chain risk, handing the Pentagon a decisive appellate win in the most consequential legal fight between an AI company and the U.S. government to date. The decision, issued by a panel of Judges Karen LeCraft Henderson, Gregory Katsas, and Neomi Rao, overturns in effect the August 28 district-court ruling that had found the Pentagon’s measures “illegal and baseless” — and it leaves the maker of Claude barred from military work for the foreseeable future.

What the court decided

At the heart of the case is the so-called supply-chain-risk designation that the Pentagon applied to Anthropic earlier this year. That label, issued under Defense Secretary Pete Hegseth, did more than exclude Anthropic from direct military contracts. Under what court filings and commentators came to call the Hegseth Directive, any contractor doing business with the U.S. military was effectively barred from doing any commercial business with Anthropic at all — a sweep that one law-firm analysis described as extraordinary in its reach. Three contractors either terminated their work with Anthropic or were instructed to do so by the department, according to findings by Judge Rita Lin of the Northern District of California, who handed Anthropic its first major win in March by preliminarily blocking the designation.

The D.C. Circuit panel saw the matter differently. Writing for the majority, Judge Gregory Katsas concluded that the Defense Department acted within its lawful authority when it designated Anthropic a security risk, and that the court was not the right place to second-guess the executive branch’s national-security judgments about who belongs in the defense industrial base. In his conclusion, Katsas acknowledged what he called the “profoundly difficult questions” raised by the case — including Secretary Hegseth’s stated concerns that an overly constrained AI policy could leave the United States at a disadvantage — but found that those concerns were policy questions for the political branches, not justiciable claims for courts to resolve.

Judge Karen LeCraft Henderson dissented. Her skepticism of the Pentagon’s case had been on display since oral argument in May, when she said she saw no evidence of any “maliciousness, malintent or sabotage” from Anthropic and pressed the government’s counsel on what, precisely, the company had done to warrant exclusion from the supply chain.

The road to this ruling

The litigation timeline is a story of dueling courts and dueling postures. Anthropic sued Defense Secretary Pete Hegseth and the department in March after being declared a supply-chain risk, arguing the designation was applied to punish the company on ideological grounds — specifically, for its public stance against allowing its technology in autonomous weapons and domestic surveillance. Judge Lin’s March 27 injunction blocked the designation, calling it “illegal and baseless,” and in September the EFF celebrated a further ruling that the department had unlawfully retaliated against Anthropic’s protected speech.

But the government kept winning at the appellate level. On April 8, the D.C. Circuit declined to block the designation while the case proceeded, and court watchers noted early on that Anthropic drew a difficult panel: Henderson, Katsas, and Rao were the same three judges who had rejected the company’s earlier request for a stay. Insider Defense observed at the time that this made a loss on the merits likely, leaving en banc review — a rehearing before the full D.C. Circuit — as the probable next step. That prediction held. Friday’s 2-1 decision, in case No. 26-1049, is the merits loss that Anthropic’s lawyers had feared.

The backdrop is unambiguous political pressure. Court documents note that one day after Anthropic’s decision that triggered the dispute, President Trump and Secretary Hegseth denounced the company on social media, and the Secretary began the process that led to the blacklist. The D.C. Circuit’s new opinion recounts that sequence in laying out how the designation came to be.

What it means for Anthropic

The immediate practical effect is that Anthropic remains shut out of federal defense work — direct contracts and, more damagingly, the web of relationships with defense primes and contractors that the Hegseth Directive chills. Anthropic’s commercial business is not directly touched by the ruling, and the company’s momentum elsewhere is considerable: Bloomberg reported this month that its annualized revenue has topped $65 billion as it moves closer to a potential IPO, and its Claude Opus 5.5 launch just this week reset frontier-model expectations. The blacklist is thus less an existential threat than a standing constraint — a ceiling on how much of the fastest-growing AI customer segment (national security) the company can serve, enforced by the very government whose contracts its rivals happily take.

The litigation is not over. The startup could petition for rehearing en banc before the full D.C. Circuit, and failing that, the Supreme Court remains the final stop. Judge Henderson’s dissent gives the company a ready-made argument that the panel was divided on questions of first impression about how much judicial scrutiny applies when the executive branch uses procurement authority against a company for its protected speech.

What it means for the industry

The broader significance extends well beyond one company. The ruling is the clearest judicial statement yet that courts will afford the executive branch wide latitude in deciding which AI companies are trustworthy enough for the defense supply chain — and that AI firms’ public safety stances can carry real commercial consequences in Washington. It effectively green-lights a tool that future administrations can wield against other AI contractors whose politics they dislike, a prospect that Judge Henderson’s dissent and the EFF’s earlier reporting both flagged as dangerous.

It also sharpens the divide among frontier labs. OpenAI, Google, and Meta have all courted defense customers aggressively; Anthropic alone took the position that cost it access, and the courts have now said the price of that position can be enforced by decree. For an industry already sorting itself into national-security insiders and outsiders, Friday’s decision makes the boundary official.

What to watch

Three things matter next. First, whether Anthropic seeks en banc review within the coming weeks, and whether the full court takes it — the D.C. Circuit has been friendly to executive power of late, and the panel’s earlier stay denial suggests limited appetite for relitigating. Second, whether the retaliation claim, which a district judge allowed to proceed in September, survives as a separate track; a win there could restore some leverage even if the designation stands. Third, the September 29 White House meeting between President Trump, Speaker Johnson, and AI company executives — a summit already on the calendar — where the Pentagon-Anthropic feud is certain to loom over the discussion of how the government and the AI industry will share power, money, and responsibility for the frontier.

For now, the blacklist stands, the dissent is on record, and the most important AI-policy court fight of the year has entered its next act.