Anthropic Pentagon rulings split; Claude dropped

A judge ruled the Anthropic Pentagon designation unlawful

On August 27, 2026, U.S. District Judge Rita Lin granted summary judgment to Anthropic in its suit against the Department of War, the name the Trump administration now uses for the Department of Defense (DoD). Her 59-page opinion and a four-page final order threw out the supply chain risk designation that Secretary of War Pete Hegseth had placed on the company.

This was not a preliminary ruling. In March the same court had paused the government's actions because Anthropic looked likely to win. This time the judge reviewed the administrative record the government filed to justify itself and decided the case on facts neither side disputed.

The court found retaliation for Anthropic's public criticism

The core finding is that the government punished Anthropic for speaking out. Judge Lin held that the actions violated the First Amendment because they were retaliation for protected speech. They also violated the Fifth Amendment, since Anthropic got no notice or chance to respond before losing its government business.

By the end of the case, the government's only reason specific to Anthropic was that it could not "trust" the company. It pointed to Anthropic's criticism of the department in the press.

The judge concluded the real goal was to make an example of Anthropic for its "arrogance," not to guard against sabotage. The government's own conduct cut against it:

Only a four-page memo backed the supply chain risk label

A supply chain risk designation under 10 U.S.C. § 3252 lets the Pentagon exclude a supplier when an adversary could use it to sabotage or subvert a national security system. The judge found the government had not shown that risk.

The whole rationale was a four-page memo, dated after two of the three actions it was meant to justify. It was written by Emil Michael, the department's under secretary for research and engineering.

The memo's main theory was that Anthropic could reach into Claude after delivery through a backdoor. The government abandoned that theory during the case and conceded that Claude poses no more national security risk than any other "black box" AI model.

The judge declared the designation arbitrary and capricious and beyond what § 3252 allows. The final order set aside three things:

• the § 3252 supply chain risk designation
• the part of Hegseth's directive that barred anyone doing business with the U.S. military from any commercial activity with Anthropic
• the cutoffs nine agencies, including State, Treasury, Homeland Security and Energy, imposed after the president's order

Anthropic did not win everything. The court rejected its claim that the president's directive violated the separation of powers, and ruled for agencies that had taken no action or only interim steps.

The Anthropic Pentagon dispute began over two uses of Claude

Anthropic was a Pentagon supplier before it was a Pentagon target. The court's account of the fight starts with a contract negotiation.

The Pentagon wanted Claude available for all lawful uses

In July 2025 the department's Chief Digital and AI Office gave Anthropic a two-year agreement worth up to $200 million. A month later, Anthropic and the General Services Administration announced a deal to offer Claude Gov across all three branches of government. Per the administrative record, the Pentagon flagged no supply chain concern about the company before March 2, 2026.

In fall 2025 the two sides began talking about putting Claude on GenAI.mil, the department's internal AI platform. The Pentagon asked Anthropic to drop every usage restriction and accept a clause letting the military and its contractors use Claude "for all lawful uses."

Anthropic gave up most of its limits but held on to two: no lethal autonomous weapons and no mass surveillance of Americans. CEO Dario Amodei told Michael by email that, based on Anthropic's training and testing, Claude was not safe for either job.

Hegseth's deadline and Trump's order on February 27

At a February 24 meeting, Hegseth praised Claude's capabilities and called Anthropic's concerns understandable. He also said plenty of other vendors had never raised them.

He gave Anthropic until 5 p.m. on February 27 to accept "all lawful uses" or be labeled a supply chain risk, and mentioned the Defense Production Act as another way to force the issue. Two days later Amodei said publicly that Anthropic could not agree in good conscience.

At 3:47 p.m. on February 27, before the deadline passed, President Trump posted on Truth Social ordering every federal agency to stop using Anthropic's technology, with six months for agencies such as the Pentagon to phase it out. He accused the company of trying to make the department follow its terms of service instead of the Constitution. Hegseth followed with a directive naming Anthropic a supply chain risk and barring the military's contractors from any business with it.

The formal § 3252 letter, dated March 3, reached Anthropic the next day. Anthropic sued on March 9, and the court froze the government's actions with a preliminary injunction on March 26.

Most Claude customers were never covered by the label

For most companies that use Claude, nothing in this fight ever applied. The measures reached federal agencies and businesses that work with the U.S. military.

Defense contractors no longer face a blanket ban

Hegseth's directive went far beyond the Pentagon's own systems: it told every contractor, supplier and partner of the military to stop doing any business with Anthropic, even work that had nothing to do with defense. That order is now vacated, and the defendants must withdraw the guidance they issued to carry it out.

Anthropic had argued from the start that a supply chain designation could not reach that far. In its March 5 statement, the company said the designation only covered Claude used as part of a contract with the Department of War. The Pentagon's other designation, under § 4713, was upheld by the D.C. Circuit on September 25, so the limits on Claude inside Pentagon contracts still stand.

The Pentagon is still free to choose another AI vendor

The ruling does not put Anthropic back into Pentagon AI work. Judge Lin wrote that the department is free to pick whichever AI supplier it wants, and the final order says it does not require the department to use Claude or stop it from moving to other providers.

What changed is how the department can walk away. It keeps every lawful option it had before February 27, including dropping Claude, but it cannot brand a U.S. company a national security threat to punish its speech.

Courts split on Anthropic and the Pentagon dropped Claude

The litigation is not over, but the two courts that heard Anthropic's challenges now disagree. On September 25, the D.C. Circuit upheld the Pentagon's second designation of Anthropic in a 2-1 decision, and ten days later a defense official told the BBC that the Pentagon has stopped using Anthropic products. The Pentagon made that choice; Anthropic did not stop offering Claude.

The D.C. Circuit upheld the § 4713 designation 2-1

Section 4713 comes from the Federal Acquisition Supply Chain Security Act (FASCSA). Challenges to a designation under it skip the district court and go straight to the D.C. Circuit, so Anthropic had to fight on two tracks. On April 8, a panel of Judges Henderson, Katsas and Rao refused to suspend the designation while the case went on, but agreed to hear it quickly.

Writing for the majority, Judge Katsas, joined by Judge Rao, found that the Pentagon had solid grounds to treat Claude as a national security risk under the statute. The opinion leaned on two facts: Anthropic builds usage limits into Claude, and more than once those limits kept the model from doing work that government users asked for. A dispute had also come up over whether the contract allowed Claude to be used in a military operation then under way overseas, which left the department unsure the model would do what it needed.

The majority turned down the constitutional claims as well. In its view, the department told Anthropic about the exclusion and its reasons promptly and gave it a fair chance to push back, and the exclusion was about Anthropic refusing contract terms, not about its speech.

Judge Henderson dissented. She read the statute's word "manipulate" narrowly, as covering only deliberate deception.

Anthropic can still ask for a rehearing

CNBC reported that the panel held off on making its ruling take effect right away. That gives Anthropic time to ask the same three judges to look at the case again, or to ask the full D.C. Circuit to hear it (known as en banc review). An Anthropic spokesperson said the company disagrees with the decision, noted that the San Francisco court had already struck down the parallel designation, and said every option is on the table.

What the government will do about the August 27 judgment is still unclear: the district court docket ends with the case closed on August 28 and shows no new notice of appeal. On September 3, Michael, who wrote the memo the San Francisco court found wanting, posted on X that Anthropic is still a designated supply chain risk for the department and the defense industrial base. Unite.AI, which reported the post, took it as a sign that the department would keep the label in public while the cases continue.

The government's appeal of the March injunction, filed with the Ninth Circuit on April 2, has been on hold while it waited for the D.C. Circuit. Before the San Francisco judgment, the law firm Jones Walker wrote that if the two appeals courts reached different answers, a Supreme Court petition would be likely. The Ninth Circuit has not ruled yet.

The Pentagon says it has stopped using Anthropic products

A defense official told the BBC in a statement on October 5 that the Pentagon no longer uses Anthropic products. Hegseth had said in February that the department would be off Anthropic by late August, so the switch landed more than a month after that target. Anthropic declined to comment on the statement.

Lauren Kahn, a former defense official who is now a senior research analyst at Georgetown's Center for Security and Emerging Technology (CSET), told the BBC the delay shows how hard these tools are to swap out. Once a model is wired into a system, she said, pulling it back out can be painful.

While the court fight went on, the Pentagon signed deals with Google, xAI and OpenAI, and two people told the BBC that OpenAI's tools have spread in some military departments. Ties with the White House may be easing, too: President Trump met Amodei twice the week before the statement. For more on where Anthropic stands as a company, see Anthropic IPO: Date, Price and What Is Confirmed.

Claude ran inside Palantir's Maven Smart System

Former defense officials and contractors who spoke to the BBC said Claude was still in use as recently as the week before the statement. By their account, it helped with research, analysis and intelligence gathering, and with military operations against Iran. None of them were named.

Claude was built into the Maven Smart System, a data platform run by Palantir that the Pentagon relies on to organize intelligence and other data. According to the BBC's sources, analysts pulled satellite images and drone video together in Maven and then passed that material to Claude and other large language models, for example to help pick out possible military targets. One person said Pentagon staff kept using Claude, including Anthropic's Mythos models, for the whole length of the dispute.