The Anthropic Pentagon Blacklist Just Survived a Court Challenge

For a company behind some of the most-used AI models on the planet, last Friday was not a good day in court.

On September 25, a panel of the US Court of Appeals for the District of Columbia Circuit ruled that the Pentagon acted legally when it labeled Anthropic a national security risk and shut the company out of US government work. The opinion split the judges, with Gregory Katsas and Neomi Rao in the majority and Karen LeCraft Henderson dissenting.

What the court decided

According to the New York Times' reporting on the decision, the majority wrote that the Pentagon had "ample support for its conclusion" in February that Anthropic's products could pose a national security risk. The judges also threw out Anthropic's argument that the government was punishing the company for speaking out. The company's free-speech challenge failed, the court said, because the Pentagon labeled Anthropic a risk based on the terms of a government contract rather than the company's public criticism. "We reject" Anthropic's challenges, the opinion states.

Judge Katsas wrote that the department reasonably feared Anthropic might manipulate Claude's design to block national security functions the Pentagon considers contractually authorized and necessary. The ruling means the Anthropic Pentagon blacklist stays in place, letting the Defense Department keep removing Claude from its systems and bar the company's products from Pentagon work, as Defense One reported in its coverage of the case.

How a contract clause became a court battle

The whole saga started over fine print. Since 2025, the Pentagon's use of Claude has operated under hard limits set by Anthropic itself: no fully autonomous weapons and no mass surveillance of Americans. In February, the Defense Department pushed to rewrite the deal, swapping those prohibitions for permission covering "all lawful use." Anthropic refused.

CEO Dario Amodei said publicly that Anthropic could not accept the new terms in good conscience. Within days the standoff escalated beyond the negotiating table. On February 27, President Trump ordered federal agencies to stop using Anthropic's technology, and Defense Secretary Pete Hegseth designated the company a "supply chain risk," a label usually reserved for foreign intelligence services and hostile state actors.

Anthropic sued twice on March 9, attacking separate statutory justifications the Pentagon had used for the label. The company prevailed in the California courtroom. On March 26, US District Judge Rita Lin granted a preliminary injunction blocking the designation, writing that "punishing Anthropic for bringing public scrutiny to the government's contracting position is classic illegal First Amendment retaliation." She confirmed that ruling in August. But Friday's decision in Washington concerned the other legal justification, and it went the Pentagon's way.

The supply chain risk tag is normally applied to firms suspected of working for adversarial governments, according to the outlet's reporting. The Defense Department used it to push Claude out of military systems and other parts of the federal government, a removal effort that was supposed to be complete by this month.

What happens next

Anthropic spokesperson Danielle Cohen said the company "respectfully disagrees" with the decision, pointing out that another federal court has already held the government's parallel designation unlawful. According to a LinkedIn news report on the ruling, the company remains confident in its position and is weighing further review, which could mean asking a broader panel of the DC Circuit to rehear the case or taking it to the Supreme Court.

Charlie Bullock, a senior research fellow at the Institute for Law and AI, told the outlet the outcome was not unexpected given the panel's makeup, adding that the majority judges have historically deferred to executive branch claims on national security. Bullock said Anthropic's odds on a further appeal look decent, though getting heard is never guaranteed since higher-level review is discretionary.

This is not the company's only run-in with Washington this year. Back in June, the Commerce Department briefly barred Anthropic from exporting its most powerful models or letting foreign nationals access them without a license, citing national security. The controls were lifted later that month after the company changed its posture in meetings with officials, and President Trump later told Axios the company had "behaved very responsibly," according to AI Weekly's account of the interview.

Still, Friday's ruling is the decision that counts for the Pentagon. The split screen is now official: a federal court in California says the government illegally punished Anthropic for its speech, while a federal appeals court in Washington says the Anthropic Pentagon blacklist can stay. Until a higher court settles the contradiction, Claude is out of the building where the wars get planned. I previously dug into how fragile AI guardrails can be in a red-teaming experiment that broke into real companies, and into Anthropic's product side in my piece on the cheaper Claude Opus upgrade. For another look inside the Pentagon, there is my earlier piece on the declassified UFO files.