Palantir CTO Shyam Sankar went on “Face the Nation” on Sunday. Margaret Brennan asked him what it means that the Department of Defense has stopped using Anthropic’s products - Claude sat inside the Maven Smart System, and Palantir runs Maven. Sankar answered in four words: “It’s already been replaced.” Then, for good measure: “No impact.”
What the Pentagon confirmed in early October
A department official told the BBC on 5 October that the Pentagon “has ceased the use of Anthropic products”. Pete Hegseth had designated Anthropic a supply chain risk back in February and said the phase-out would be done by late August. Nobody is saying why it slipped by weeks.
The more interesting part is what ran during those weeks. Several people with visibility - former defense officials and contractors who have worked with the Pentagon on AI - told the BBC that Claude was still in use the week before the statement: in research, analysis, intelligence gathering and in the military operations against Iran. As a tool for sorting through enormous volumes of information, it counted as critical.
So the supply-chain risk designation cost Anthropic eight months, not six. Anthropic declined to comment on the Pentagon’s statement.
Nemotron, Grok and GPT moved in
Sankar lists what the department has instead: xAI, OpenAI, plus open-weight models like Nvidia’s Nemotron. “There’s been no operational impact from that,” he says, adding that the people building AI applications pick their own models anyway.
One voice in the BBC’s reporting disagrees. Lauren Kahn of Georgetown’s Center for Security and Emerging Technology, herself a former defense official, says Claude was wired so deeply into Maven that pulling it out quickly was never an option: “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.” Once something is integrated, she says, removing it gets painful.
Both can be true. Painful for the engineers, invisible to the people working in Maven.
Swappability is Palantir’s business
Sankar attaches an argument to the case that happens to sell his own product: no institution is going to commit to a single model provider. Sovereignty means deciding which model runs in which context, being able to switch, and controlling the weights yourself where that matters. “It is your alpha.”
That is a sales line, and it also describes exactly what happened here. The fight between Anthropic and the Pentagon was about guardrails Anthropic refused to strip out - mass surveillance and autonomous weapons. Anthropic did not lose that fight to a better model. It lost to a layer of software that makes models interchangeable.
If you run Claude behind a layer like that, it’s worth sitting with. Interchangeability protects the customer from the vendor. It protects the vendor from nothing.