After months of a high-stakes public and legal conflict over access and safety protocols, the U.S. Department of Defence has confirmed it has fully ended all use of AI tools developed by leading artificial intelligence firm Anthropic, multiple sources close to the matter have confirmed.
The standoff between the Pentagon and Anthropic dates back to February of this year, when Defence Secretary Pete Hegseth publicly labeled the company a national security supply chain risk and set an end-of-August deadline for the full removal of its tools from all department systems. Despite that public order, multiple sources including former senior defence officials, AI defence contractors and department insiders told the BBC that Anthropic’s flagship Claude model remained in active use across critical Pentagon operations as recently as last week.
For more than two years starting in 2024, Anthropic held a groundbreaking position in U.S. government AI deployment: it was the first advanced AI company to have its models cleared for use in classified work across federal agencies, and Claude quickly became an indispensable tool for the Pentagon. The model was integrated into the Maven Smart System, the Palantir-operated core platform the U.S. military uses to organize intelligence data, drone footage, satellite imagery and operational information. Defence analysts relied on Claude to sort through massive datasets, identify potential military targets, and condense complex intelligence assessments into executive briefs for senior command, even during active military operations against Iran, according to insiders.
The root of the conflict between the two sides emerged earlier this year, when Pentagon officials pushed Anthropic to remove built-in AI safety guardrails from Claude and grant the military unrestricted, unfettered access to the model’s full capabilities. Anthropic leadership refused the demand, citing long-standing company concerns about the risks of mass surveillance and unregulated development of autonomous lethal weapons systems. In response, Hegseth applied the national security supply chain risk designation to Anthropic—a classification historically reserved for companies based in adversarial foreign nations.
Anthropic pushed back forcefully against the designation, calling it “unprecedented and unlawful” and filing a lawsuit against the Trump administration to overturn the labeling. The case remains ongoing in federal courts.
Even as the legal battle proceeded, the Pentagon moved to replace Anthropic’s tools, signing new AI contracts with other major industry players including Google, Elon Musk’s xAI, and OpenAI. In recent months, OpenAI’s models have seen expanded adoption across multiple military departments, two insiders familiar with the department’s AI rollout confirmed.
Despite the public order to phase out Anthropic by late August, full removal took far longer than initially planned, industry analysts note, because of Claude’s deep integration into the Maven platform. Lauren Kahn, senior research analyst at Georgetown University’s Center for Security and Emerging Technology and a former U.S. defence official, explained that integrated AI tools are not simple plug-and-play systems that can be swapped out quickly. “Once they become integrated it can be painful to remove them,” Kahn noted, adding that the months-long timeline for full removal underscores the technical complexity of overhauling embedded AI infrastructure in large government systems.
Interestingly, the Pentagon’s confirmation of the full phase-out comes amid a surprising warming of relations between Anthropic and the White House. Just last week, Anthropic CEO Dario Amodei met with President Donald Trump twice at the White House, alongside other leading technology executives. Speaking to reporters after one meeting with Trump at his side, Amodei acknowledged that AI risk remains a key concern for the company, but expressed optimism about collaboration. “How we address those risks is still under discussion,” Amodei said, “but if we work with the president, we can win safely.” Trump also offered public praise for the Anthropic CEO following the meeting, calling him “great” and “fantastic.”
A spokesperson for Anthropic declined to provide any comment on the Pentagon’s phase-out announcement when contacted by the BBC. Details of the full termination of the department’s use of Anthropic products have not been reported prior to this confirmation.