Pentagon Ends Use of Anthropic AI After Months of Internal Resistance and Legal Battle

The US Department of Defence has confirmed it has stopped using artificial intelligence tools made by Anthropic, according to a statement provided to the BBC, bringing an end to a contentious chapter that pitted the military's operational needs against the company's safety policies.

A department official said the Pentagon "has ceased the use of Anthropic products," a disclosure that follows months of reporting suggesting the tools remained deeply embedded in military workflows well after political leaders ordered them removed.

The development marks the culmination of a dispute that began earlier this year when Defence Secretary Pete Hegseth designated Anthropic a supply chain risk on national security grounds, a classification typically reserved for companies headquartered in countries considered hostile to the United States.

Hegseth announced in February that the Pentagon would phase out Anthropic's technology by late August, setting a six-month deadline that fell just one day before the United States and Israel launched military operations against Iran.

Yet multiple people familiar with the matter told the BBC that as recently as last week, Anthropic's Claude model was still being used across the department for research, analysis and intelligence gathering, as well as in support of military operations against Iran.

Those sources, including former US defence officials and contractors who have worked closely with the Pentagon on artificial intelligence, described Claude as a critical tool for processing enormous volumes of information. The details of the department's full withdrawal from Anthropic's products had not previously been reported.

Anthropic declined to comment when approached about the Pentagon's statement. The company had been supplying AI tools to the US government and military since 2024 and was the first advanced AI developer to have its technology deployed in government agencies conducting classified work.

The breakdown in the relationship stemmed from the Pentagon's request earlier this year that Anthropic remove safety guardrails from Claude and grant the military unrestricted access to its capabilities. Anthropic refused, citing concerns about mass surveillance and the potential development of autonomous weapons.

In response, Hegseth applied the national security supply-chain risk designation, a legal mechanism that Anthropic described as "unprecedented and unlawful." The company has since sued the Trump administration in an effort to have the designation overturned.

The legal fight has unfolded alongside the Pentagon's efforts to diversify its AI suppliers. In recent months, the department has signed contracts with several major technology companies, including Google, xAI and OpenAI. According to two people familiar with AI use at the defence department, OpenAI's tools have been adopted more widely across some military departments during this period.

The Pentagon's statement on Monday followed multiple interviews in which sources said Claude remained in widespread use in recent weeks. According to those accounts, the model was embedded within a larger data platform known as Maven Smart System, which is operated by Palantir and serves as the department's primary tool for organising intelligence and other data.

Lauren Kahn, a senior research analyst at Georgetown's Center for Security and Emerging Technology, said Claude's deep integration into Maven would have made a rapid removal difficult. "The fact that it took the Pentagon until now to remove all use of Claude from across the system is indicative that these things are not just plug and play," she said, adding that once such tools become integrated, "it can be painful to remove them." Kahn is herself a former US defence official.

The BBC spoke with defence contractors specialising in AI who have worked closely with the Pentagon, as well as former defence officials and others familiar with the department's use of artificial intelligence. Several spoke on condition of anonymity because they were not authorised to speak to the media.

One former senior US defence official recalled that when Hegseth first ordered Anthropic to be phased out, "there was concern with moving beyond Anthropic to OpenAI" and other available tools. Despite the directive, Pentagon employees continued to use Claude, including its powerful Mythos models, "throughout the whole period of the controversy," according to one person with knowledge of the situation.

Claude was employed in particular for intelligence gathering and analysis, including through Palantir's Maven system, several sources said. Palantir, a leading defence technology firm, was founded by billionaire Peter Thiel, a prominent ally of President Donald Trump.

Sources familiar with how the technology is used explained that analysts and other defence officials rely on Maven to access and organise visual data, including satellite imagery and drone footage. That data was then fed into Claude and other large language models for tasks such as identifying potential military targets.

Officials frequently used the technology when compiling one-page briefs that were passed up the chain of command, according to a source who was given a demonstration of Maven by Palantir earlier this year.

The episode highlights the growing tension between the rapid integration of AI into military operations and the safety policies of the companies that develop the technology. It also underscores the practical challenges facing governments that seek to unwind such dependencies once established, particularly when the tools are woven into broader platforms used for time-sensitive national security work.