Pentagon moves past Anthropic for a multi‑vendor AI backbone

Anthropic’s Claude used to run inside Palantir’s Maven Smart System, a platform the Pentagon relies on. After the Defense Department labeled Anthropic a supply chain risk, Claude was pulled out.

In its place, the department now uses models from OpenAI, Google and xAI, along with open-weight models, which anyone can download and run, such as Nvidia’s Nemotron. Sankar said the people building AI applications inside the military choose which model to use for each job.

That flexibility is good business for Palantir. Sankar said the ability to pick and swap models is what has been driving the company’s commercial growth. Instead of betting on one AI supplier, Palantir sells the layer that lets customers plug in whichever one suits them.

The dispute broke out in February. Anthropic refused to remove restrictions that stop its models being used for mass domestic surveillance or fully autonomous lethal weapons. The Pentagon’s view was that any lawful military use should be allowed, and that it isn’t a supplier’s job to set the rules for how the armed forces work.

President Donald Trump then told federal agencies to stop using Anthropic’s products. Defense Secretary Pete Hegseth designated the company a supply chain risk under two separate laws, with the formal designation dated March, and barred military contractors from working with it.

Anthropic says AI is not yet reliable enough to be used safely in autonomous weapons, and that it opposes domestic surveillance as a violation of basic rights. Trump and Hegseth accused the company of putting American lives at risk with what they called ideological posturing.

The legal battle has gone both ways. In August, US District Judge Rita Lin in San Francisco struck down one of the designations, finding the administration had unlawfully punished Anthropic for its views on AI safety. She also blocked the government-wide ban and the order stopping contractors from doing business with the company.

Then, in late September, a three-judge federal appeals panel in Washington sided with the Pentagon by two votes to one on the other designation. Judge Gregory Katsas, joined by Judge Neomi Rao, said the department reasonably feared Anthropic could shape Claude to refuse tasks the military considers necessary. The judges said they didn’t doubt Anthropic’s good intentions, but the test turned on what the company does rather than why. Judge Karen LeCraft Henderson dissented.

That ruling lets the Pentagon keep removing Claude and keep contractors from using it for defense work. It does not bar Anthropic from the federal government as a whole. Pentagon spokesman Sean Parnell said the decision completely validates the department’s position. Anthropic said it respectfully disagrees and, in October, asked the court to rehear the case. The company says the blacklisting has cost it billions of dollars in lost business, and its stock market prospectus lists the fight as a risk to government sales.

Sankar went further than the contract dispute. He blamed effective altruism, a movement focused on large long-term risks, for much of the public fear around AI, calling it a “doomsday cult”. He also claimed that many AI researchers have taken up smoking to signal their belief in AI, but named no one and gave no evidence.

On rules, he argued that existing agencies such as the FDA should regulate how AI is applied in their areas, while saying the country will still need a system for reporting and responding to AI incidents. Palantir didn’t sign last month’s White House AI accord, which Sankar said was meant for companies that build AI models. He left the question of who should run incident reporting to the new AI czar, Jay Clayton.

For ordinary users, nothing changes in the apps you use day to day. Claude is still available to the public, and OpenAI, Google and xAI keep their consumer products too. The bigger shift is about who sets the limits. The Pentagon’s position is that a supplier shouldn’t build in rules that could stop a model from doing a lawful job. Anthropic’s position is that some uses are too risky no matter who asks.

The rivalry is now personal as well as commercial. OpenAI struck its own deal with the Pentagon and says it includes explicit bans on mass domestic surveillance, directing autonomous weapons and high-stakes automated decisions. Anthropic president Daniela Amodei has publicly asked whether OpenAI’s move was a principled stand or an opportunistic one.

What to watch next: whether the full appeals court agrees to rehear Anthropic’s case, and whether any other AI company draws a line the Pentagon won’t accept.