The Pentagon Says It’s Stopped Using Claude, But Sources Say It Was Still Running Last Week

A Pentagon official told the BBC the department has stopped using Anthropic's Claude, capping an eight-month phaseout ordered by Trump. But sources say it was still running in Pentagon systems, including Iran-related operations, as…

Get a summary in:
Aerial view of the Pentagon building
Follow Teck Hustlers on Google
Add Teck Hustlers as a preferred source to see more of our stories in Google Top Stories.
Add as preferred source on Google
AAAdmin AlexWriter
MAM.AliFact-Checker

Published: October 5, 2026 · Last updated: October 5, 2026

TL;DR: A Pentagon official told the BBC that the Department of Defense has “ceased the use of Anthropic products,” closing out an eight-month standoff that began when President Trump ordered federal agencies to drop the company in February. But BBC sources describe Claude still running inside Pentagon systems, including intelligence work tied to the Iran conflict, as recently as last week.

Server racks in a data center used for classified AI workloads

The Pentagon just announced it has finished pulling Claude out of its systems. A Pentagon official told the BBC on Monday that the department “has ceased the use of Anthropic products,” officially closing out a fight that started in February and dragged on for the better part of a year.

The dispute goes back to February 27, when President Trump publicly directed federal agencies to stop using Anthropic’s products, and Defense Secretary Pete Hegseth designated the company a supply chain risk for Pentagon contracts. The fight was never really about how well Claude performed. It was about what Anthropic would and wouldn’t let the military do with it.

A day earlier, Dario Amodei had published a statement saying Anthropic “cannot in good conscience” accept a Pentagon contract term requiring support for “any lawful use” of its models. Anthropic drew the line at two things it said it would never support: mass domestic surveillance and fully autonomous weapons that make lethal decisions with no human involved. The Pentagon didn’t take the refusal well, and the supply chain designation followed almost immediately.

Hegseth set a six month phaseout window, originally expected to wrap up around August. It didn’t. A DefenseScoop report on September 11 said the department was still “poised” to move all classified AI workloads off Anthropic by October. Two weeks later, on September 25, a federal appeals court rejected Anthropic’s legal challenge to the supply chain designation, with Judge Karen LeCraft Henderson dissenting. That ruling cleared the last real obstacle between the Pentagon and total removal.

Here’s where the official version runs into a problem. Sources who spoke to the BBC, described only as former defense officials and contractors, say Claude was still running inside Pentagon systems as recently as last week, including in military operations connected to the Iran conflict. They say it had been used for research, analysis, and intelligence work, and that it was embedded directly in Palantir’s Maven Smart System, the platform the Pentagon uses to organize intelligence data across operations.

That gap between “we’ve stopped” and “it was still running last week” isn’t necessarily a cover-up. It’s closer to what happens when you build something complicated on top of a specific AI model and then try to pull it back out. Lauren Kahn, a senior research analyst at Georgetown’s Center for Security and Emerging Technology and a former defense official, told the BBC the part everyone in this story keeps dancing around: “Once they become integrated it can be painful to remove them.” The Pentagon’s statement doesn’t say when the final access was actually cut, and nothing in the reporting says how many users, systems, or contractor environments were still touching Claude the day that statement went out.

None of this was a scramble for a brand new vendor. The Pentagon had already awarded contracts to all four major model makers, Anthropic, Google, OpenAI, and xAI, back in July 2025, so shifting classified workloads meant routing around Anthropic rather than negotiating a new deal from nothing. Commercial access to Claude, outside Pentagon contract work, is unaffected by any of this.

Related: Anthropic’s relationship with the rest of the tech world hasn’t been one-sided lately. Amazon shut out Meta’s AI agent and then opened its seller tools to Claude just weeks before this fight went public, and the Pentagon’s appetite for outside tech talent hasn’t cooled either, with Elon Musk and Palmer Luckey now co-leading Project Meridian, its 120-day study on what the military needs next.

Bottom Line: You can order a company out of the building, but the wiring stays put until somebody actually rips it out. The Pentagon got its political win on paper eight months after Trump demanded it, and by its own sources’ account, kept using the thing it banned for most of the time it took to actually make that true.

Follow Teck Hustlers: Facebook · X · Instagram · LinkedIn · TikTok · Pinterest

Follow Teck Hustlers on Google
Add Teck Hustlers as a preferred source to see more of our stories in Google Top Stories.
Add as preferred source on Google