The Pentagon Said It Cut Off Anthropic. For Months, Claude Was Still Running Inside Its War Room.
A February order gave the Pentagon six months to pull Claude from military systems over a supply chain risk designation. The deadline passed at the end of August. Sources say Claude stayed embedded in the Pentagon's Maven intelligence platform anyway, and the Pentagon didn't say otherwise until a reporter asked.
By Mara Voss, Technology
· 3 min read · Updated
Key Takeaways
- •The Pentagon designated Anthropic a national-security supply chain risk on February 27, triggering a six-month deadline to remove Claude from military systems.
- •That deadline passed at the end of August. The Pentagon did not confirm the removal was complete until October 5, after the BBC asked.
- •Sources told the BBC that Claude remained active inside the Pentagon's Maven intelligence platform in the meantime, including work tied to operations against Iran.
- •Anthropic lost one of two related court fights: a D.C. Circuit panel upheld the designation 2-1 on September 25, a week after a San Francisco judge had ruled a separate justification unlawful.
The Pentagon told the BBC on October 5 that it has "ceased the use of Anthropic products." The ban was supposed to start in August. Multiple people with direct knowledge told the BBC that Claude was still running inside a Pentagon intelligence platform as recently as last week.
The short answer
The Pentagon designated Anthropic a national-security supply chain risk on February 27, giving itself six months to pull Claude from military systems. That deadline passed at the end of August. The Pentagon did not confirm the removal was finished until October 5, after the BBC asked, and sources say Claude stayed embedded in the Pentagon's Maven intelligence platform in the meantime, including work tied to operations against Iran.
A ban that outlasted its own deadline
The friction started in September 2025. Anthropic had signed a $200 million Pentagon contract two months earlier, and the two sides were negotiating how Claude would run on the Defense Department's GenAI.mil platform. The Pentagon wanted unrestricted access to the models for any lawful purpose. Anthropic wanted a written guarantee that Claude would not be used for fully autonomous weapons or domestic mass surveillance. Neither side moved. The deal collapsed.
$200 million
Pentagon contract Anthropic signed in July 2025
Talks over military access to Claude collapsed two months later.
Defense Secretary Pete Hegseth responded by accusing Anthropic of trying "to seize veto power over the operational decisions of the United States military." In February he labeled the company a supply chain risk, a designation Washington has mostly used against firms tied to adversary governments, not an American AI lab that already held a Pentagon contract.
Pentagon staff kept using Claude anyway. Former defense officials and contractors told the BBC that Claude's Mythos models stayed active inside Maven, the Palantir-built platform the Pentagon uses to organize satellite imagery, drone footage, and other intelligence, for research, analysis, and military operations against Iran, "throughout the whole period of the controversy."
“Once they become integrated, it can be painful to remove them.”
Two lawsuits, one win apiece
Anthropic sued over the designation in two courts at once, because the Pentagon used two separate legal justifications that had to be fought separately. A federal judge in San Francisco ruled on August 27 that one of the two was unlawful. A panel on the U.S. Court of Appeals for the D.C. Circuit went the other way on September 25, voting 2-1 to uphold the second justification.
Judge Gregory Katsas, writing for himself and Judge Neomi Rao, both Trump appointees, sided with the Pentagon's reading of the Supply Chain Security Act: "In our Republic, it is the President and the Secretary of War who must determine how best to balance the competing risks." Judge Karen LeCraft Henderson, appointed by George H.W. Bush, dissented. The available reporting does not detail her reasoning.
Anthropic said it respectfully disagreed and is weighing a rehearing request or a petition to the Supreme Court. A company spokesperson declined to comment when the BBC asked about the Pentagon's October 5 statement.
Who is filling the gap
The Pentagon has since signed contracts with Google, xAI, and OpenAI. Two people familiar with AI use inside the department told the BBC that OpenAI's tools have picked up more of the work in some military divisions over the past several months.
The legal fight has not stopped Anthropic CEO Dario Amodei from warming up to the White House. Trump met with him twice last week. After one meeting, Amodei told reporters he still has concerns about AI risk, but that "if we work with the president, we can win safely." Trump called him "great" and "fantastic" standing right next to him, a reversal from a Truth Social post three weeks earlier that mocked Amodei as a "perfect little angel."
The gap between the two statements is the story. A supply chain risk designation sounds like a kill switch. In practice, it took a government press office seven months, one lost contract, two lawsuits, and a reporter's question to actually flip it.
- Anthropic
- Pentagon
- Claude
- Maven
- defense contracts
- AI policy
Sources
Corrections
No corrections have been made to this article.
About the reporter
Technology Reporter, Trestlewire
I spent seven years as a product manager at a mid-size SaaS company before I ever wrote a sentence for pay, which means I have sat through more roadmap reviews than most people would tolerate in a lifetime. I watched a scheduling feature get rebranded three times before it shipped, and I watched a launch date slide past four straight quarters while the slide deck stayed exactly the same. That is where the question I still ask every day came from: does this actually ship, or is it a demo.
Read full bio and all stories →