The U.S. military has formally designated AI firm Anthropic a supply chain risk, the company announced Thursday, a step that could bar it from military-related contracts. Anthropic is currently the only AI provider deployed on the Pentagon’s classified networks.
The dispute centers on guardrails Anthropic wants written into contracts for its Claude model. Anthropic has sought explicit prohibitions preventing the U.S. military from using Claude for mass domestic surveillance or to power fully autonomous weapons. The Pentagon has resisted, saying it must be able to use Claude for “all lawful purposes” and maintaining that the uses Anthropic fears are already illegal or limited by Defense Department policy.
Defense Secretary Pete Hegseth publicly said last week Anthropic would be cut off from government contracts and designated a supply chain risk. A senior Pentagon official told CBS News the company only received formal notification this week. Hegseth said the military would phase out Anthropic over six months; the official said the formal designation itself did not specify a removal timeline.
Two sources previously told CBS News the military used Claude in strikes on Iran that began last weekend, though how the model was used has not been made clear.
Anthropic CEO Dario Amodei said the company believes the designation is legally unsound and intends to challenge it in court. He added that most commercial customers will be unaffected, arguing the designation should only apply to uses of Claude tied directly to Defense Department contracts and should not prevent military contractors from using the technology for nonmilitary work.
The designation followed Amodei’s statements that he remained in talks with the Pentagon to “deescalate the situation.” He told investors and spoke at a Morgan Stanley conference that the company and the government “have much more in common than we have differences.” In interviews Amodei reiterated Anthropic’s two red lines — no mass domestic surveillance and no fully autonomous weapons — saying AI could give the government surveillance powers “contrary to American values” and that AI is not yet reliable enough to safely select targets without human oversight.
The Pentagon’s position is that mass surveillance of Americans is illegal and that fully autonomous weapons are already constrained by internal policy, so it should not be required to accept additional written restrictions. Emil Michael, the Pentagon’s chief technology officer, said last week that while “at some level, you have to trust your military to do the right thing,” the department cannot agree to be unable to defend itself “in writing to a company.” The Pentagon offered to acknowledge in writing the laws and policies that restrict surveillance and autonomous weapons; Anthropic said that compromise was inadequate, warning legal wording could allow the military to sidestep the guardrails.
Relations between the administration and Anthropic turned sharply public, with heated rhetoric on both sides. Hegseth called the company “sanctimonious,” Michael accused Amodei of a “God-complex,” and former President Trump labeled Anthropic “radical left” and “woke.” With no agreement reached, the administration ordered federal agencies to stop using Claude immediately and gave the Defense Department up to six months to phase the technology out. Rival OpenAI said it reached a deal with the military.
A senior Pentagon official described the dispute as “about one fundamental principle: the military being able to use technology for all lawful purposes,” and said the department will not let a vendor “insert itself into the chain of command by restricting the lawful use of a critical capability.” Amodei called the administration’s action “retaliatory and punitive,” saying Anthropic’s stance was taken “for the sake of this country” and defending the company’s decision as patriotic: “Disagreeing with the government is the most American thing in the world.”