Product1 publisher2 min readPublished
Pentagon says it has stopped using Anthropic's Claude months after labelling it a supply-chain risk
Pentagon told the BBC it has ceased using Anthropic's products, months after Pete Hegseth gave the department six months to drop them. For federal AI buyers, it shows a vendor designation being carried out even when the tool sits deep inside daily intelligence work.
The Product Desk · Product desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Hegseth's designation came after Anthropic refused to strip Claude's safety guardrails over mass-surveillance and autonomous-weapons concerns, under a law usually aimed at firms from hostile countries.
- People familiar with the matter told the BBC that Claude was still in use as recently as last week for research, analysis, intelligence gathering and operations against Iran.
- Claude ran inside the Maven Smart System, the Palantir-operated platform the Pentagon uses as its main tool for organising intelligence and other data.
- The Pentagon has meanwhile signed contracts with Google, xAI and OpenAI.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- precedent A US AI vendor has now been designated and dropped over a dispute about what its model may be used for. Any supplier whose usage limits clash with a defence customer's demands carries the same risk.
- constraint Anthropic's lawsuit calling the designation "unprecedented and unlawful" has not kept Claude in Pentagon use. Vendors cannot treat a pending court case as protection while a phase-out runs.
- contradiction The cutoff is attested only by a department statement, with Anthropic declining to comment, so buyers should treat Claude's removal as the Pentagon's claim until an outside source confirms it.
Inside the Maven Smart System, analysts pull up satellite imagery and drone footage. Claude and other large language models then work through that material, with uses that include picking out possible military targets, according to people familiar with how the platform is used [18]. One of them told the BBC that Pentagon employees kept using Claude, including its Mythos models, "throughout the whole period of the controversy" [3].
Hegseth set the phase-out deadline on 27 February, one day before the US and Israel attacked Iran [17]. Six months from that date is about 27 August [19]. For the Pentagon staff using it, Claude was a critical tool for sorting through vast quantities of information [2]. When the phase-out was first ordered, "there was concern with moving beyond Anthropic to OpenAI" and other tools, a former senior US defence official told the BBC [16].
Users keep working in the platform they open every morning, with whatever model is wired into it, until someone takes it out. Claude had been in government use since 2024, and Anthropic was the first advanced AI company to have its tools deployed in agencies doing classified work [7]. Two people familiar with AI use in the department said OpenAI tools have been more widely adopted in some military departments in recent months [11].
Lauren Kahn, a senior research analyst at Georgetown's Center for Security and Emerging Technology and a former US defence official, said Claude's deep integration into Maven would have made it difficult to remove quickly [13]. "The fact that it took the Pentagon until now to remove all use of Claude from across the system is indicative that these things are not just plug and play," she said [14]. "Once they become integrated it can be painful to remove them," Kahn said [15].
A buyer working under federal rules can sort each AI tool on two lines. One is whether staff call the model directly or reach it inside a platform a contractor runs. The other is whether a second model is already wired into the same workflow. A direct tool with a fallback in place is the cheapest to unwind if its vendor is barred. An embedded tool with no fallback is the most expensive, because the removal waits on another company's engineering work. Maven sat between the two. Palantir runs it [12], and it already fed data to other large language models [18]. The department still needed months after the designation before an official said Claude was gone [4].
What to watch
- A court ruling in Anthropic's suit to overturn the supply-chain-risk designation.
- Confirmation from Palantir or people inside the department that Claude is out of Maven, or fresh reports that analysts are still using it.
- Which model replaces Claude inside Maven for intelligence and targeting work.