Palantir report on Anthropic’s military hesitancy sparked rift with Pentagon, sources say

The escalating tension between the Pentagon and AI startup Anthropic was reportedly triggered by Palantir, a key collaborator with both the company and the US Department of Defense.

According to a report by Semafor, Anthropic stands as one of the few “frontier” large language models authorized for classified US government use, made accessible through Amazon’s Top Secret Cloud and Palantir’s Artificial Intelligence Platform (AIP).

This integration led to the Claude chatbot appearing on the screens of officials monitoring the movements of then-Venezuelan President Nicolás Maduro. The tracking of Maduro and the potential for military intervention reignited a wave of activism within Silicon Valley regarding the deployment of proprietary tech products in US government operations.

Shortly after the operations involving Maduro, a Palantir executive held a routine check-in with an Anthropic official. During this dialogue, the Anthropic representative reportedly discussed the operation with the Palantir executive. Following the exchange, the Palantir official concluded that Anthropic did not approve of its Claude technology being utilized for such tactical purposes.

Concerned by the implication that Anthropic might resist the application of its technology in US military operations, the Palantir executive reported the conversation to the Pentagon. A senior Defense Department official confirmed this sequence of events to Semafor.

According to several individuals briefed on the matter, this report led to a significant breakdown in Anthropic’s relationship with the Pentagon. Semafor previously reported that Defense Secretary Pete Hegseth criticized Anthropic in a speech on January 12. Referring implicitly to the company, Hegseth stated, “We will not use AI models that you cannot use to fight.”

An Anthropic spokesperson characterized the account of the meeting between the company and Palantir as “inaccurate.” The spokesperson asserted that the company “has not discussed this matter or expressed concerns with any industry partner outside of routine discussions on purely technical issues.”

“Anthropic is committed to utilizing frontier AI to support US national security,” the spokesperson continued. “To this end, we were the first frontier AI company to deploy our models on classified networks and the first to offer customized models for national security customers. Claude is used across the government, including the Department of Defense, in a wide range of intelligence-related use cases, in accordance with our Usage Policy.”

However, Anthropic has reportedly declined to sign an “all legal uses” contract with the Pentagon—an agreement that would ensure Claude’s unrestricted deployment. Sources close to the matter indicate that Anthropic is seeking specific exemptions that would prohibit certain types of surveillance and the development of autonomous weapons. These sources suggest that the relationship between Anthropic and the Pentagon has deteriorated significantly since these negotiations stalled.

A Defense Department official suggested to Semafor that the military is beginning to lose confidence in Anthropic, viewing its models as a potential “supply chain risk.” The official further indicated that there have been vague threats to bar subcontractors, such as Palantir, from utilizing Anthropic’s technology.

Such a formal designation, though rarely invoked by the Pentagon, could alienate private sector clients and threaten Anthropic’s business prospects as the company prepares for an initial public offering (IPO) later this year.

Behind the scenes, the two parties continue to negotiate contract terms. “We are engaged in productive, good-faith discussions with the DoW [Department of War] on how to continue this work and resolve these complex issues,” the Anthropic spokesperson said.

In May, Semafor reported that Anthropic had drawn the ire of the Trump administration by lobbying against legislation that would prevent states from adopting their own AI regulations. While Anthropic claims to be seeking to avoid “fragmented” laws that could slow innovation, critics have accused CEO and co-founder Dario Amodei of a cynical attempt at regulatory capture to stifle competition.

White House AI Lead David Sacks described the move as a “sophisticated regulatory capture strategy based on fear-mongering,” a charge Anthropic denies. Anthropic also faced White House scrutiny in September after refusing requests from companies working with federal law enforcement, citing a refusal to grant exceptions for tasks involving the surveillance of US citizens.

Military planners now express concern over relying on vendors like Palantir if there is a risk that a critical technology component could be withdrawn mid-operation due to a violation of the supplier’s usage policies. Military officials speaking to Semafor emphasized their belief that suppliers should have no role in determining how the Pentagon employs technology during active operations.

Conversely, Semafor technology editor Reed Albergotti suggested that Claude did not play a specialized role in the Maduro incident. “I spoke with employees at Palantir about how exactly the company uses frontier language models like Claude, and the reality is that these models typically represent only about 10% to 20% of any customized software application,” Albergotti noted. “We are still far from the Joint Chiefs creating a ‘General Claude’ chatbot to run the military.”

As Albergotti explained, Palantir develops bespoke software solutions and typically utilizes language models to make these programs more accessible or to assist with specific, open-ended data mining tasks that aid strategic analysis.