The Pentagon's Anthropic Ban Just Collapsed: Palantir Is Ignoring It Publicly

Palantir CEO Alex Karp went on live television and told CNBC that Palantir is still using Anthropic's Claude. The Pentagon's response: silence.

"Our products are integrated with Anthropic, and in the future, it will probably be integrated with other large language models," Karp said Thursday at the company's AIPcon 9 conference. He was not hedging or qualified or speaking in past tense. He was describing the current state of Palantir's operations.

Palantir is the closest thing the Pentagon has to a principal AI contractor. The company has classified contracts running through every branch of the military. Palantir's systems are embedded in the decision-making infrastructure of U.S. defense operations. If anyone has the leverage to defy the Pentagon's "supply chain risk" designation without consequence, it is Palantir.

And on Thursday, Palantir proved exactly that.

Karp's statement is not a minor vendor preference. It is a direct contradiction of the Pentagon's stated policy that Anthropic represents an unacceptable supply-chain risk. It is an admission that the Pentagon's most critical defense contractor cannot operationally comply with the Pentagon's own directive. And it is a legal gift to Anthropic's lawyers, who are preparing to argue that the Pentagon's action is unenforceable and unconstitutional.

What Karp actually said

"Our products are integrated with Anthropic," Karp said. Not "were integrated." Present tense. Current state. Active reliance.

"In the future, it will probably be integrated with other large language models," Karp continued. This is not a statement about migrating away from Anthropic. This is a statement about adding other models while keeping Anthropic. Diversification, not replacement.

The implication is unambiguous: Palantir is still using Claude, will continue using Claude, and does not plan to remove Claude from its systems.

That statement was made in public, on a recorded conference call, to a major financial news outlet, by the CEO of the company most embedded in Pentagon operations. If Karp had wanted to hide his company's defiance, he would not have announced it to CNBC.

The Pentagon's admission of defeat

On the same day Karp made his statement, the Pentagon's Chief Technology Officer Emil Michael said something equally damaging to the government's case: "You can't just rip out a system that's deeply embedded overnight."

Let that statement settle. The Pentagon's own CTO just admitted that the Pentagon cannot actually remove Anthropic from its systems because Anthropic is too deeply integrated.

That is not a temporary obstacle. That is not a transition challenge. That is an acknowledgment that the Pentagon's supply-chain risk designation is unenforceable against major contractors because Claude is too essential to Pentagon operations.

Palantir's public defiance combined with the Pentagon CTO's public admission of inability to enforce the ban creates a legal catastrophe for the government's position in the March 24 TRO hearing.

Anthropic's lawyers will cite both statements in their briefs. They will argue: "Your Honor, the Pentagon itself has admitted it cannot enforce this designation. The largest defense contractor is publicly ignoring it. The supply-chain risk argument is not about actual risk — it is about political retaliation."

The enforceability question

The Pentagon's "supply chain risk" designation was supposed to be a procurement tool. In theory: the government identifies a vendor as a supply-chain risk, and contractors remove that vendor from their systems within a specified timeframe (six months).

In practice: Palantir is the largest government contractor the Pentagon has. If Palantir does not comply, who will the Pentagon punish? Fire Palantir? Break the company? Palantir is the Pentagon's AI infrastructure. The Pentagon cannot afford to punish Palantir for using Claude any more than it can afford to lose Palantir's services entirely.

The Pentagon's leverage is asymmetrical and broken. It can fire Anthropic. It cannot fire Palantir.

Palantir, understanding this dynamic perfectly, publicly announced it will not comply.

Why this matters for the lawsuit

Anthropic filed suit arguing that the Pentagon's supply-chain risk designation violates due process, the Administrative Procedure Act, and First Amendment principles. The company argued that the designation is retaliation for refusing to grant "any lawful use" to Claude.

The Pentagon's defense was supposed to be: "This is a standard vendor procurement decision made through ordinary channels for legitimate national security reasons."

Karp just shattered that defense. By publicly announcing that Palantir still uses Claude, Karp demonstrated that the supply-chain risk designation is not about actual risk. It is about political punishment of a company the Pentagon disagreed with.

If the risk were real, Palantir would have to remove Claude. If Palantir cannot remove Claude without operationally destroying its own systems, then the risk cannot be as severe as the Pentagon claimed.

Michael's admission that "you can't just rip out a system that's deeply embedded" is even more damaging. It is the Pentagon's CTO saying, on the record, that the Pentagon's own policy is operationally impossible to implement.

The Palantir-Anthropic relationship

Palantir and Anthropic partnered with Amazon Web Services in 2024 to offer joint solutions to the Department of Defense. The partnership was announced publicly, with Pentagon blessing. The three companies built systems specifically designed to serve military operations.

That partnership is what made Anthropic essential. Palantir did not randomly choose to integrate Claude. Anthropic integrated Claude into Palantir's systems as part of a government-approved vendor partnership. The Pentagon knew about the integration. The Pentagon signed off on the integration.

Then, when Anthropic refused the Pentagon's "any lawful use" demand, the Pentagon punished Anthropic by issuing the supply-chain risk designation. But the same designation threatened to break Palantir's own operations.

Palantir had two choices: comply with the Pentagon and destroy its own products, or defy the Pentagon and keep its products operational. Karp chose to keep his products operational.

What happens next

The Pentagon will not publicly punish Palantir for defying the Anthropic ban. The damage would be too great. Losing Palantir's operational support would cripple Pentagon technology operations far more than losing Anthropic would.

So Palantir's defiance will stand. Palantir will continue using Claude. The Pentagon CTO's admission that the ban is unenforceable will stand in the public record.

Anthropic's legal team will cite both of these in the March 24 hearing. The judge will read Karp's statement. The judge will read Michael's statement. The judge will understand what has happened: the government tried to use a procurement tool to punish a private company, but the government's own infrastructure depends on that company, so the tool does not work.

The legal implication: if the ban cannot be enforced, the ban cannot stand in court.

The enterprise consequence

For enterprise AI buyers watching this: the Pentagon's most important defense contractor just demonstrated that it does not have to comply with Pentagon AI policy.

If Palantir can use Claude despite the Pentagon ban, so can you.

The government's leverage against you is even weaker than it is against Palantir. You are not a strategic defense contractor. The Pentagon cannot threaten to break your company for defying them the way it would have to with Palantir.

You can use Claude. You can defy the Pentagon's supply-chain risk designation. The worst the government can do is remove you from government contracts. That might matter or might not, depending on your business model.

What Palantir proved is that the Pentagon's AI policy is not actually enforceable against vendors that integrated Claude deeply into their operations. And for most enterprises, that integration has not happened yet. You have even more freedom to choose than Palantir does.

The geopolitical angle

While the Pentagon was fighting with Anthropic over policy, China's government was subsidizing OpenClaw development in Shenzhen and Wuxi. By the time the Pentagon settled its dispute with Anthropic, China had already moved on to the next generation of AI infrastructure.

Palantir's public defiance of the Anthropic ban is a win for Anthropic in the short term. But it is also evidence that the Pentagon spent months fighting an internal battle while external competitors moved forward.

The Pentagon's AI strategy should be coherent and forward-focused. Instead, it is spending energy on retaliation against a vendor while the only company that can actually execute the retaliation (Palantir) is openly ignoring the policy.

That incoherence is the real national security risk.

What you should understand

The Pentagon cannot actually remove Anthropic from the defense industrial base because the defense industrial base is too dependent on Anthropic now. The supply-chain risk designation was supposed to be a threat. It turned into a bluff.

Palantir called the bluff publicly.

That has three immediate consequences:

First, Anthropic's legal position in the March 24 hearing just improved dramatically. The government's own contractor just demonstrated that the ban is unenforceable.

Second, enterprise AI buyers have permission to use Anthropic without fear of government retaliation. If the Pentagon's largest contractor can ignore the ban, you can ignore it too.

Third, the Pentagon's AI policy credibility is damaged. If the Pentagon says "X is a supply-chain risk," but the Pentagon's own contractors ignore that designation without consequence, then the Pentagon does not actually have a supply-chain risk policy. It has a suggestion.

For vendors: that means government procurement decisions are not binding on enterprise procurement. You can win government contracts and lose enterprise contracts based on policy disagreements that have nothing to do with technical capability.

For buyers: that means you have leverage. The government told you not to use Anthropic. Palantir just told you the government cannot enforce that directive. You are free to make your own choice.

---

Frequently Asked Questions

Q: Could the Pentagon actually fire Palantir for defying the Anthropic ban?

A: Legally, yes. Practically, no. Palantir is too critical to Pentagon operations. Firing Palantir would cripple Pentagon technology infrastructure more than losing Anthropic would. The Pentagon will not make that choice. Palantir knows this. That is why Karp could make his statement without consequence.

Q: Is this the death of the Pentagon's Anthropic ban?

A: In practical terms, yes. The largest defense contractor just announced publicly that it will not comply. The Pentagon's CTO admitted the policy is unenforceable. In court, this will be cited as evidence that the ban is not about actual supply-chain risk, but about political retaliation. The legal death sentence comes March 24 when the judge reads Karp's statement.

Q: What does this mean for government AI policy going forward?

A: It means the government cannot use procurement as a weapon against vendors if those vendors are too integrated into government operations. The government has leverage over new vendors but not existing vendors. For enterprise buyers: pick the vendor you think is best. The government's procurement decisions are not binding on private decisions.