AINEWS
Anthropic, the US AI lab behind the Claude model, refused to drop safety rules for military use. A federal judge had already called the underlying blacklist retaliatory.
The Pentagon has "ceased the use of Anthropic products," a Department of Defense official told the BBC on Monday. The reason is that Anthropic would not strip the safety guardrails from its Claude model for military use.
Anthropic's concerns, on the record since February, were mass surveillance and autonomous weapons. Defense Secretary Pete Hegseth had demanded unrestricted military access to the model. When Anthropic refused, Hegseth declared the company a "supply-chain risk to national security," a legal label usually reserved for firms based in countries that pose a threat to the United States. Anthropic called the move "unprecedented and unlawful." The original declaration came in an X post reported by CBS News, which noted the label had not previously been applied to a US-based AI company.
A Pentagon directive, documented by Just Security, instructed components to begin phasing Anthropic out of classified systems. The company sued. On August 28, a federal judge ruled that the Trump administration had illegally retaliated against Anthropic for refusing to drop the guardrails. The ruling did not vacate the supply-chain-risk designation itself, which remains in active litigation. What it established is that the procurement lever was used as punishment.
A Defense Department official, identified only as "Michael," told DefenseScoop in September that about 90 percent of AI workloads had transitioned off Anthropic, and that all work on Maven Smart Systems, the Pentagon's flagship AI platform for intelligence synthesis and targeting, integrated through the defense contractor Palantir, had been moved months earlier. DefenseScoop reported at the time that the department was poised to move all classified AI workloads off Anthropic by October. The Pentagon has since signed new AI contracts with Google, xAI, and OpenAI. OpenAI's tools are now more widely used inside the department than Claude was at its peak.
Claude was still in active use as recently as last week. According to the BBC, the model was being used at the department for research, analysis, intelligence gathering, and military operations against Iran, with multiple people familiar with the matter saying Claude remained deployed even as the transition proceeded. Former US defense officials and contractors who have worked with the Pentagon on AI told the BBC the cutover was more abrupt in some offices than the official figures suggest, with research and analytic workloads the slowest to migrate. Anthropic was the first advanced AI company to have its tools deployed in classified government work, in use by the US military since 2024. Monday's confirmation closes a window that was still partly open.
The new contracts went to the vendors that did not push back on safety. A federal finding now says one builder was punished for refusing to do the same. Anthropic declined to comment on the Pentagon's Monday statement, and there is no public Pentagon press release accompanying the BBC report. The litigation over the underlying designation is still active. The next operational test is whether the August 28 retaliation ruling produces any procurement consequence for the contracts that absorbed the work, or whether it makes the next principled stand survivable rather than fatal.