The U.S. Department of Defense said it has stopped using Anthropic products, including the company's Claude artificial intelligence models. The move comes eight months after the department designated Anthropic as a supply-chain security risk, effectively blacklisting it on national security grounds in February.
The Pentagon recently halted its use of Anthropic products, the BBC reported on the 5th. The exact date of the cutoff is not known, but the department was still using Anthropic's Claude models as recently as last week. When Defense Secretary Pete Hegseth announced the blacklisting in February, he said the department would stop using Anthropic products by the end of August, meaning the actual cutoff came two months behind schedule.
The delay is believed to stem from technical difficulties. According to sources, Claude models had been embedded in the Maven Smart System, a large-scale data platform used by the department. Operated by Palantir, Maven is a core platform the Pentagon uses to organize intelligence and data. Lauren Kahn, a senior research analyst at Georgetown University's Center for Security and Emerging Technology, said Claude was so deeply integrated into Maven that removing it quickly would have been difficult.
Anthropic's AI products had been used by the U.S. government and military since 2024, the first case of AI being deployed at government agencies handling classified work. But earlier this year, when the Pentagon pressed Anthropic to remove Claude's safeguards and give the military unrestricted access to its AI products, the company refused, citing concerns over mass surveillance and autonomous weapons.
Anthropic objected that the demand was unlawful and filed suit, but the federal appeals court in Washington, D.C., rejected its request last month and sided with the government. While the legal fight was under way, the Pentagon signed contracts with other major technology companies including Google, xAI and OpenAI as it moved to phase out Anthropic.
Warnings Persist That Humans Are Likely to Lose Control of AI
Amid the conflict with the government, warnings about the risks of AI have continued. Jacob Coxson, a former Anthropic researcher, told a New York City Council hearing on AI on the 5th that "we don't yet know how to control any AI system." He said that "on the current path, it is more likely than not that humanity loses control of AI, and that could end in human extinction." He testified at the request of Council Speaker Julie Menin, who is pushing a package of AI regulation bills.
Coxson called the industry's approach "extremely reckless," criticizing the longstanding Silicon Valley mantra of moving fast and breaking things. "'Move fast and break things' may work for a photo-sharing app, but it does not work for building the most powerful technology in history," he said.
He also said "some kind of speed limit is needed in frontier model development," and noted that an AI "kill switch," or emergency shutoff mechanism, under review by the New York City Council could help in the short term but would not be enough to ensure safety over the long run. When Coxson left Anthropic last month, he said the company and OpenAI were "gambling with our lives."
Alex Turner, a former Google DeepMind researcher, said in video testimony that "China is not the only potential adversary. There is a fairly high probability that we are racing to build our own adversary at home — an unaligned AI," adding that "this could one day be more powerful than China."
With the Trump administration largely maintaining a hands-off stance on AI regulation, state and local governments including New York are accelerating their own rules, widening the gap between federal and local approaches.
