The Pentagon says it has stopped using Anthropic’s products, but the delay shows how hard it is to pull a language model out of a live intelligence system. The ban worked on paper in February. It took until October to work in practice.
Washington can ban an AI vendor in one social media post. Unplugging that vendor from a working military platform takes far longer.
Key Takeaways
- The Pentagon told the BBC on Monday that it has stopped using Anthropic’s products.
- Defense Secretary Pete Hegseth set a six-month phase-out on 27 February, so the department finished roughly five weeks late.
- Anthropic refused to drop limits on mass domestic surveillance and fully autonomous weapons.
- Sources say Claude ran inside Palantir’s Maven Smart System, which explains the slow exit.
- Google, xAI, and OpenAI now hold Pentagon contracts, and OpenAI usage reportedly grew in some branches.
- Anthropic did not comment, and the usage details rest on anonymous sources.
When Did the Pentagon Stop Using Claude?
The department told the BBC on Monday that it has ceased the use of Anthropic products. That statement came only after the BBC asked about reports of continued use.
Hegseth announced the ban on 27 February and gave the Pentagon six months to move to another provider. By my count, that deadline fell in late August. The department missed it by about five weeks, and nobody has explained why.
Meanwhile, people close to the work say Claude still supported research, analysis, and intelligence gathering as recently as last week. Some of them say it also supported operations against Iran.
Why Did the Pentagon Label Anthropic a Supply-Chain Risk?
The fight started over guardrails. Anthropic had served the US government since 2024, and the BBC reports it was the first frontier lab deployed in agencies doing classified work. Then the Pentagon pushed for unrestricted access, and Anthropic said no.
The company drew two lines: mass domestic surveillance and fully autonomous weapons. Hegseth answered with a supply-chain-risk label, a tool that normally targets firms tied to hostile countries. Anthropic’s own statement called the move unprecedented for an American company.
Hours after the designation, OpenAI announced a classified-environment deal with the Pentagon. Anthropic then sued over the designation. The BBC reports that a judge has since ruled the administration retaliated illegally.
How Did Claude End Up Inside Maven Smart System?
Maven is the Pentagon’s main platform for organizing intelligence, and Palantir runs it. Sources told the BBC that Claude sat inside it. Analysts pulled satellite imagery and drone footage through Maven, then fed the data to language models.
Those models helped with tasks like drafting the one-page briefs that travel up the chain of command. Former officials say the work included identifying potential targets.
Lauren Kahn, a Georgetown analyst and former defense official, told the BBC these systems are not plug-and-play. Integration is the whole story.
Think of Claude as the translator in a meeting. Swap the translator and the whole room changes its pace, its phrasing, and its trust in the output. Nobody can fix that with a deadline.
The calendar made it worse. Hegseth’s clock started one day before the US and Israel attacked Iran. The department had to rebuild part of its intelligence workflow in the middle of a war.
Which AI Vendors Filled the Gap Anthropic Left?
Several labs now hold Pentagon contracts, and each took a different deal. The table below shows what the reporting confirms.
| Vendor | Pentagon status | Terms on record |
|---|---|---|
| Anthropic | Designated supply-chain risk on 27 February; formal letter followed on 4 March | Refused to drop limits on mass domestic surveillance and fully autonomous weapons |
| OpenAI | Announced classified-environment deal hours after the designation | Usage reportedly growing in some military branches |
| Classified-network agreement signed in late April | “Any lawful government purpose” language | |
| xAI | Contract signed, per BBC | Terms not reported |
Google’s deal deserves a closer look. Reporting on Google’s classified-network deal indicates the company cannot audit how its models run on air-gapped systems. Anthropic wanted contractual limits on that kind of use. Google accepted the government’s definition of “lawful.”
What Does the Delay Mean for AI Companies?
Usage policy now carries procurement risk. Anthropic held its lines and lost a major customer. Every other lab will weigh that cost before it draws a line of its own.
Switching costs also beat benchmarks. A model wired into a platform like Maven earns loyalty through habit, not leaderboard rank. That favors any vendor already embedded, and it makes late entrants fight uphill.
Legal wins arrive slowly, though. Even if Anthropic prevails in court, the Pentagon has already rebuilt around other vendors.
The politics also refuse to line up. Trump met Dario Amodei twice at the White House last week and praised him in public. The same administration spent months trying to phase his company out of the military. One government, three postures: punish, court, and depend.
FAQs
Q. Has the Pentagon stopped using Claude?
Yes, according to a department official’s statement to the BBC. Nobody has independently verified it, and Anthropic declined to comment.
Q. Why did the Pentagon ban Anthropic?
Anthropic refused to remove safeguards against mass domestic surveillance and fully autonomous weapons. Hegseth then labeled it a supply-chain risk.
Q. What is Maven Smart System?
Maven is the Pentagon’s primary platform for organizing intelligence and other data. Palantir operates it.
Q. Which AI companies replaced Anthropic?
The Pentagon has signed contracts with Google, xAI, and OpenAI. Two sources told the BBC that OpenAI tools gained ground in some military departments.
Q. Did a court overturn the designation?
The BBC reports a judge ruled the administration retaliated illegally. The reporting does not say the designation has been reversed.
Q. Does the ban affect Anthropic’s other customers?
Anthropic says no. It argues the designation covers only Claude use on Department of War contracts, not how contractors serve other customers.
Related: GPT-6 Astra vs Claude Fable 5.1: Which AI Model Is Better in 2026?
