Anthropic refuses Pentagon demands for unrestricted AI access

On Thursday, Anthropic CEO Dario Amodei formally refused to provide the Pentagon with unrestricted access to the company’s artificial intelligence systems. The refusal sets up a direct confrontation with the US Department of Defense (referred to in negotiations as the Department of War) ahead of a Friday 5:01 p.m. deadline. With less than 24 hours remaining until the deadline set by Defense Secretary Pete Hegseth, Amodei stated that he cannot in good conscience accede to the request. He explained that while Anthropic understands that the military makes military decisions, the company believes that in a narrow set of cases, AI can undermine, rather than defend, democratic values.

The core dispute centers on Anthropic’s refusal to allow its Claude model to be used for the mass surveillance of Americans or in fully autonomous weapons. According to an Anthropic spokesperson, overnight contract language from the Department of War made virtually no progress on preventing Claude’s use for mass surveillance of Americans or in fully autonomous weapons. The spokesperson added that new language framed as a compromise was paired with legalese that would allow those safeguards to be disregarded at will, noting that these narrow safeguards have been the crux of negotiations for months. In response to the impasse, the Department of Defense has threatened to either label Anthropic a supply chain risk or invoke the Defense Production Act—a US law that gives the president authority to force private companies to prioritize production for national defense. Amodei pointed out the contradiction in these threats, noting that one labels the company a security risk while the other treats Claude as essential to national security.

While the Department of Defense is reportedly preparing competitor xAI for military use, Anthropic is currently the only frontier AI lab with classified-ready systems for the military. Amodei expressed a strong preference to continue serving the department and US warfighters, provided the two requested safeguards remain in place. However, if the government decides to offboard the company, Amodei stated that Anthropic will work to enable a smooth transition to another provider to avoid disrupting critical operations. Summarizing the firm’s stance on the potential split, Amodei, the CEO of Anthropic, said: “We can just part ways. There’s no need to be nasty about it.”

Why it matters

This standoff highlights the growing friction between frontier AI labs and national defense agencies over the ethical boundaries of military AI, with significant implications for how governments regulate and procure dual-use technology.