Last week, tensions between the Department of Defense and artificial intelligence giant Anthropic reached a boiling point.
Anthropic, the developer of the Claude chatbot system and a frontier AI company with up to $200 million worth of defense contracts under its belt, has built its brand around promoting AI safety and touts the lines it won’t cross.
Now, the Pentagon appears to be pushing the envelope.
Hints of a possible rift between Anthropic and the Department of Defense (now renamed the Department of the Army) began to grow after the Wall Street Journal and Axios reported that Anthropic products were used in the operation to capture Venezuelan President Nicolas Maduro.
It is unclear how Anthropic’s Claude was used.
Anthropic has not raised or found any policy violations following Operation Maduro, according to two people familiar with the matter, who requested anonymity to discuss a sensitive topic. It is said that there is high visibility into how the company’s AI tool “Claude” is being used in data analysis operations.
Anthropic was the first AI company to be allowed to provide services on a classified network through its partnership with Palantir in 2024. In announcing the partnership, Palantir said Claude could be used to “support government operations, including rapidly processing vast amounts of complex data” and “help U.S. authorities make more informed decisions in time-sensitive situations.”
Palantir is one of the military’s preferred data and software contractors, offering services such as collecting data from space sensors to improve targets for soldiers. His activities under the Trump administration and law enforcement agencies have also drawn scrutiny.
Anthropic has maintained that it does not and will not allow its AI systems to be used directly for autonomous lethal weapons or domestic surveillance, but concerns were raised by Anthropic employees after reports of its technology being used in connection with the Venezuela attack through a contract with Palantir.
Semafor reported Tuesday that during a regular meeting between Anthropic and Palantir, Palantir executives were concerned that Anthropic employees appeared to disagree about how the company’s systems were used in the operation, which could lead to “a rupture in Anthropic’s relationship with the Department of Defense.”
A senior Pentagon official told NBC News that “Anthropic executives have contacted Palantir executives to ask if their software was used in the Maduro attack.”
A Pentagon official said Palantir executives were “alarmed that the question was posed in a way that suggested Anthropic might not have authorized its software to be used during the attack.”
An Anthropic spokesperson would not confirm or deny that the company’s Claude chatbot system was used in Operation Maduro, citing the sensitivity of military operations. “We cannot comment on whether Claude or any other AI model has been used in any specific operation, sensitive or otherwise,” a spokesperson told NBC News in a statement.
A spokesperson denied the idea that the incident caused any significant repercussions, telling NBC News that the company had no unusual discussions with its partners about using the Claude and did not share any mission-related disagreements with the military.
“Antropic does not consult with the Department of the Army regarding the use of the Claude for specific operations,” the spokesperson said. “Additionally, we have not discussed this matter or raised any concerns with our industry partners outside of day-to-day discussions about strictly technical issues.”
Palantir did not respond to a request for comment.
The core tension between Anthropic and the Department of Defense appears to be rooted in a broader conflict over the future use of Anthropic’s systems by the military. The Department of Defense has recently emphasized that it wants to be able to use all available AI systems for any purpose permitted by law, but Anthropic has said it wants to maintain its own guardrails.
“The Department of the Army’s relationship with Anthropic is being reviewed,” Pentagon Chief Press Secretary Sean Parnell told NBC News.
“Our country calls on our partners to be willing to help combatants win any battle,” he said in a statement.
“At the end of the day, this is about the safety of our military and our people.” On Tuesday, Undersecretary of Defense Emil Michael said the department’s negotiations with Anthropic had stalled due to disagreements over the potential use of its system, according to CNBC.
In early January, Secretary of Defense Pete Hegseth released a new AI strategy document, calling for all contracts with AI companies to remove company-specific guardrails and restrictions on how the military uses a company’s AI systems, and newly allowing “all lawful uses” of AI for Department of Defense purposes.
The document calls on defense officials to incorporate this language into the Pentagon’s AI contracts within 180 days, which would involve Anthropic’s dealings with the military.
Anthropic has broadly supported the use of its services for national security purposes, but has maintained that its systems would not be used for domestic surveillance or fully autonomous weapons.
The Department of Defense has balked at Anthropic’s claims on these two issues and is increasing pressure on the company.
“Claude is used in a wide range of intelligence-related applications across the government, including the Department of the Army, consistent with our usage policy,” an Anthropic spokesperson said. “We are in good faith and productive discussions with the Department of the Army about how to continue our work and resolve these complex issues.”
Compared to other AI companies, Anthropic has prioritized enterprise and national security applications for its AI systems. In August 2025, Anthropic established a National Security and Public Sector Advisory Board comprised of former senior defense and intelligence officials, and last week added President Donald Trump’s former chief of staff Chris Liddell to its board of directors.
Anthropic will partner with Palantir starting in late 2024 to provide U.S. defense and intelligence agencies with access to a variety of Claude systems. At the time, Anthropic’s head of sales and partnerships, Kate Earle Jensen, said the company was “proud to be at the forefront of bringing responsible AI solutions to America’s classified environments and enhancing analytical capabilities and operational efficiencies in critical government operations.”
Anthropic, along with other major U.S. AI companies such as OpenAI and Google, signed a separate two-year contract with the Department of Defense in July 2025. The contracts are worth up to $200 million each to support “prototype frontier AI capabilities to advance U.S. national security.”
“Anthropic is committed to leveraging frontier AI to support U.S. national security,” an Anthropic spokesperson told NBC News in a statement. “We were the first frontier AI company to deploy our models on a classified network, and the first to offer customized models for national security customers.”
Anthropic CEO Dario Amodei has consistently emphasized Anthropic’s commitment to using AI services for national security purposes. “Democracies have a legitimate interest in some military and geopolitical tools powered by AI,” Amodei wrote in an essay published in late January. “Democracies should be equipped with AI, but only with caution and within limits.”
Michael Horowitz, who leads AI and emerging technology policy at the Pentagon and is now a political science professor at the University of Pennsylvania, said any concerns about using Antropic’s systems for active involvement in lethal autonomous weapons are likely irrelevant to the current negotiations, given the type of systems Antropic is developing.
“I would be surprised if a human model would be suitable for use in an autonomous lethal weapons system at this point, because that algorithm would be more bespoke than Claude’s,” Horowitz told NBC News.
“My sense is that Anthropic wants to expand the depth and scope of its collaboration with the Department of Defense. As far as we know, this seems to be more of a dispute over theoretical possibilities than an actual use case.”
