The US Department of Defense is no longer using Anthropic's AI tools, an official told the BBC on Monday. The department "has ceased the use of Anthropic products", the statement said — months after it designated the company a supply chain risk on national security grounds.
Defence secretary Pete Hegseth applied that label in February and said the Pentagon would stop using Anthropic by late August. The deadline passed. People familiar with the matter told the BBC that as recently as last week, Claude was still being used for research, analysis and intelligence gathering — and in military operations against Iran.
Why a ban took eight months to carry out
The explanation is not bureaucratic foot-dragging, and it is what makes this more than a procurement footnote.
Claude was not a subscription somebody could cancel. It was embedded inside the Maven Smart System, the Pentagon's primary platform for organising intelligence and other data, which is operated by Palantir. The model sat underneath a system other systems depend on.
Lauren Kahn of Georgetown's Center for Security and Emerging Technology, a former defence official, put it directly: because Claude was deeply integrated into Maven it could not be pulled out quickly. "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. "Once they become integrated it can be painful to remove them."
That is a procurement lesson with nothing to do with politics. An organisation that adopts a model at the application layer can switch vendors in an afternoon. An organisation that lets a model become the reasoning layer inside its core platform has acquired a dependency, and the cost of that dependency only becomes visible when it tries to leave.
How the dispute started
The Pentagon pressed Anthropic to remove safety guardrails from Claude and grant the military unfettered access. The company refused, citing mass surveillance and autonomous weapons. Hegseth then designated it a "national security supply-chain risk" — a label normally reserved for firms based in countries hostile to the United States, and never before applied to an American company.
Anthropic called the move "unprecedented and unlawful" and sued. The court fight, and what a federal appeals court made of it, is set out in the ruling on the blacklist.
The timing in the record is striking. Hegseth set the six-month phase-out on 27 February — one day before the US and Israel attacked Iran. Through the months that followed, the tool the department had formally banned was reportedly still being used in operations against that country.
Who filled the gap
While the case proceeded, the Pentagon signed contracts with Google, xAI and OpenAI, and OpenAI's tools have been more widely adopted across military departments in recent months. Anthropic had been the first advanced AI company to have its tools deployed in US government agencies doing classified work, from 2024.
One detail from a former senior defence official is worth noting: when the phase-out was ordered, "there was concern with moving beyond Anthropic to OpenAI". The concern was not loyalty. Replacing a model means re-testing everything built on top of it, and nobody inside a system wants that work.
What a buyer here should take from it
Bangladesh is not procuring frontier models for classified work, but the structure of the problem transfers exactly, and the sums are smaller only in money.
A bank or a hospital that wires a specific vendor's model into its core workflow — not as a feature but as the thing other features call — is making a decision that is cheap today and expensive to reverse. The reversal cost is never quoted in the proposal.
Three questions make it visible before signing. Can the model be swapped for another without rewriting the application that calls it? Is there a written description of what the system does if the vendor becomes unavailable, for any reason including a political one? And has anyone actually tested the swap, rather than assuming an interface guarantees it?
If the US Department of Defense needed eight months and a named platform operator to answer those questions in practice, a mid-sized Dhaka enterprise should not assume it will be faster.
The wider policy picture is in the Super Intelligence Force piece.
