Palantir prepares to support other AI models after Anthropic's clash with the Pentagon

Palantir is preparing to reduce its dependence on Anthropic amid the AI startup's conflict with the Pentagon. Palantir CEO Alex Karp said the company's products are already integrated with Claude, but will likely gain support for other large language models in the future.

Why the risk emerged

Donald Trump's administration designated Anthropic a supply chain risk and cut the company off from government contracts. The trigger was a dispute over exactly how the military may use Claude. Anthropic insisted on two restrictions: the model must not be used for mass surveillance of Americans, and must not be used in fully autonomous weapons systems. The company then sued the government in an attempt to challenge the decision.

For Palantir, this is not an abstract political quarrel. Through its partnership with Anthropic, Claude is already built into the company's products used by government and defense customers. According to Reuters, the Pentagon argued in negotiations that commercial AI can be used for any lawful purpose, even if that conflicts with the vendor's internal policy. Against this backdrop, any conflict around a single model automatically becomes a risk for the contractors that rely on it.

What Palantir says

On CNBC, Karp made it clear that there is no abrupt break. In his words, the Pentagon is only planning to phase out Anthropic gradually, the process is not yet complete, and Palantir's products still work with Claude.

"Our products are integrated with

Anthropic, and in the future will likely be integrated with other large language models."

The wording matters for two reasons. First, Palantir is not signaling an immediate break with Anthropic and is not saying that Claude is disappearing from its stack right now. Second, the company is effectively confirming a move toward a multi-model architecture. For a defense contractor, that is a logical step: if one supplier comes under political, contractual or regulatory pressure, the platform must be able to switch quickly to an alternative without rebuilding the product.

What changes for the market

On March 11, Reuters reported that the Pentagon does allow rare exceptions to the ban. An internal memo dated March 6 says Anthropic's tools may still be used after the announced six-month wind-down, but only in "rare and extraordinary circumstances" involving critical national security tasks where no alternative exists. The very existence of such caveats shows that removing one model from a defense environment is much harder than announcing it publicly.

• The ban applies not only to the Pentagon but also to contractors
• Exceptions require a separate risk mitigation plan
• Contracting officers were given 30 days to notify vendors
• The deadline for confirming full compliance is 180 days
• The priority is removing Anthropic products from systems used for nuclear and missile defense missions

For the market, this is also a sign of power shifting between model developers and infrastructure players. Anthropic's value to government customers used to lie in the fact that Claude was one of the few frontier models cleared for use in classified environments through partnerships such as Palantir's. Now the emphasis is shifting. What matters is not only the quality of the model itself, but also the platform's ability to survive a conflict between an AI vendor and the state.

Palantir is in a strong position here. The company is not obliged to bet on a single developer and can turn support for multiple LLMs into a competitive advantage. The stricter the customer's requirements for compatibility, substitution and control over the supply chain, the more valuable the intermediary that assembles different models into one working environment.

What this means

This story is not about an immediate break between Palantir and Anthropic, but about a change in the logic of the entire defense AI market. Military customers want to depend less on political disputes between vendors and to get interchangeable models within a single platform. For Palantir, it is a chance to establish itself not as a showcase for one partner, but as a universal layer through which the state can quickly switch LLMs to fit its own rules.