In the context of the militarization of AI between China and the United States, the BBC learned that the Pentagon has officially stopped using the Anthropic tool

An official told the BBC that the US Department of Defense no longer uses Anthropic's artificial intelligence (AI) tools. Months ago, the Pentagon classified the company as a supply chain risk, citing national security concerns.

The Pentagon "has stopped using Anthropic products," the Pentagon official said in a statement on Monday (October 5).

Defense Secretary Hegseth classified Anthropic as a supply chain risk in February and announced that the Pentagon would decommission its products by late August. It's unclear why the delay occurred.

Multiple sources familiar with the matter told the BBC that until last week, the department was still using Anthropic's Claude model in research, analysis, intelligence gathering and military operations against Iran.

Claude is seen as a key tool in sorting through vast amounts of information, people familiar with the matter said. They include former U.S. defense officials and contractors who have worked closely with the Pentagon on artificial intelligence.

The BBC reached out to Anthropic for comment on the Pentagon's statement, and a spokesperson for the company declined to comment.

Details of the Department of Defense’s sweeping shutdown of Anthropic tools have not been previously reported.

Anthropic's products have been used by the U.S. government and military since 2024, and it is also the first advanced AI company to have its tools deployed to classified government agencies.

But earlier this year, the Pentagon pressured the company to remove Claude’s safety guardrails and give the military unfettered access to its AI products.

Anthropic declined, citing concerns about mass surveillance and autonomous weapons. Hegseth then classified the company as a "national security supply chain risk" - a legal maneuver typically reserved for companies headquartered in countries that pose a threat to the United States.

Anthropic called the move "unprecedented and illegal" and has sued the Trump administration in an attempt to overturn the designation.

As the legal battle progresses, the Pentagon has signed contracts with other big tech companies, including Google, xAI and OpenAI.

OpenAI's tools have become more widely used in some military branches in recent months, according to two people familiar with the Department of Defense's use of AI.

The Pentagon's statement on Monday was in response to questions from the BBC about its use of Anthropic products. Several previous interviewees stated that Claude is still widely used.

Multiple sources said Claude was embedded into the Maven Smart System, a larger data platform used by the Pentagon.

Maven, operated by Palantir, is the Pentagon's primary platform for integrating intelligence and other data.

Lauren Kahn, senior research analyst at Georgetown University's Center for Security and Emerging Technology Research, said that because Claude is deeply embedded in Maven, it will not be easy to quickly remove it from the system.

"The fact that the Pentagon has only now removed Claude from the entire system shows that these systems are not plug-and-play," she said.

“Once they’re integrated, it’s going to be painful to divest them of,” added Kahn, a former U.S. defense official.

The BBC spoke to defense contractors who specialize in AI and work closely with the Pentagon, as well as former US defense officials and others familiar with the Pentagon's use of artificial intelligence. Several of them requested anonymity because they were not authorized to speak to the media.

A former senior U.S. Department of Defense official said that when Hegseth initially ordered Anthropic to be phased out across the government, "there were concerns about whether the transition from Anthropic to OpenAI" and other tools would go smoothly.

On February 27, Hegseth issued a six-month deadline to phase out Anthropic, and the next day the United States and Israel launched attacks on Iran.

Pentagon employees have been using Anthropic's Claude tools, including its powerful Mythos model, "throughout the entire dispute" with the company, a person familiar with the matter said.

Multiple sources said Claude was used specifically for intelligence gathering and analysis, including through Palantir's Maven system. Palantir is a leading defense technology company founded by Peter Thiel, a billionaire ally of President Donald Trump.

Sources familiar with how the technology is used say analysts and other defense officials rely on Maven to obtain and organize visual data, including satellite imagery and drone footage.

This data is then fed into Claude and other large language models for tasks such as identifying potential military targets.

A source who was given a demo of Maven by Palantir earlier this year said officials often use the technology when writing one-page briefings that go up the chain of command.

Some U.S. allies are also using Maven to share intelligence with the Trump administration.

Anthropic's conflict with the Pentagon comes as the company's relationship with the White House appears to be warming. Trump met Anthropic chief Amodei twice at the White House last week.

After one of those meetings with Trump and other tech leaders, Amodei said he remained concerned about the risks posed by AI.

"How to deal with these risks is still being discussed," Amodei told reporters at the White House, with Trump standing beside him, "but if we work with the president, we can win safely."

Trump later praised Amodei, calling him "amazing" and "terrific."

Background: The AI arms race, China’s pace and impact

BBC Chinese reporter Chen Yan

The Pentagon's suspension of Anthropic does not mean that the US military has withdrawn from the AI military race. At the heart of the dispute is who has the right to decide how technology is used, between the military and technology companies. Anthropic refused to abandon restrictions on large-scale domestic surveillance and fully autonomous weapons, and the U.S. Department of Defense demanded that it accept "any lawful use."

Nor is this an AI company opposed to military use versus the military. Anthropic stated that it supports the use of AI for legitimate foreign intelligence and counterintelligence missions. Sources told the BBC in February that Anthropic's large model Claude was used in the US operation to capture Venezuelan President Nicolás Maduro.

In contrast, the role of AI in the war in Iran is clearer. Palantir's head of UK and Europe, Louis Mosley, said in an exclusive interview with the BBC in April that its Maven Smart System "played a role" in combat management.

According to reports, the Maven intelligent system can collect intelligence, satellite images and drone images, analyze and propose target recommendations, and also recommend the level of force to be used based on available troops and equipment. And Claude is one of the large AI models that supports this system.

Although Palantir emphasized that the final decision still rests with people. However, interviewed experts told the BBC that once speed and strike scale become priorities, military personnel may not have enough time to verify targets, and may even rely too much on software judgment. The controversy therefore goes beyond "whether someone pressed the button", but whether the operator can truly understand and review the AI's suggestions.

The broad background of this controversy is still the competition with China. The National Security Commission on Artificial Intelligence has opposed a complete ban on AI-driven autonomous weapons and warned that the United States may lose its military technological advantage if China takes the lead in deploying AI-driven swarm drone systems. Critics believe that such claims could encourage an irresponsible arms race.

In addition to competition, China and the United States have also drawn a common bottom line. In November 2024, Xi Jinping and then US President Joe Biden reached a consensus: the decision to use nuclear weapons must be controlled by humans and cannot be left to AI. But this is not a ban on all military AI, nor does it clarify the boundaries of the use of conventional weapons, intelligence analysis, and target identification.

In Trump's second term, whether even the above consensus still exists has been questioned. Although relations between the two men have improved, Trump has repeatedly made clear that he does not intend to impose any restrictions on AI development.

What the Anthropic incident highlights is the blurring of the boundaries of militarization of AI: governments demand greater usage rights in the name of dealing with competitors; companies try to maintain restrictions. Even if the final decision is still made by humans, how to prevent humans from becoming the "approvers" of AI recommendations in high-speed decision-making is still an unavoidable issue for the military and technology suppliers.