The dispute lays bare a widening rift between AI companies that have written ethical constraints into their products and a Pentagon that wants unconditional access to commercial AI as the U.S. military accelerates autonomous systems development to compete with China.
The Pentagon has designated AI company Anthropic a supply chain risk — effectively cutting it off from defense contracts — after months of negotiations over use of its Claude chatbot in autonomous weapons systems broke down, a top Defense Department official said Friday.
Defense Undersecretary Emil Michael, the Pentagon’s chief technology officer, described the standoff publicly for the first time in a podcast appearance, saying talks with Anthropic CEO Dario Amodei collapsed over the company’s refusal to allow unrestricted military use of Claude.
“I need a reliable, steady partner that gives me something, that’ll work with me on autonomous, because someday it’ll be real and we’re starting to see earlier versions of that,” Michael said. “I need someone who’s not going to wig out in the middle.”
Anthropic said it would sue over the supply chain risk designation, which also affects its business partnerships with other military contractors. Trump separately ordered all federal agencies to immediately stop using Claude, though he gave the Pentagon six months to phase out a product reportedly embedded in classified military systems, including those used in the ongoing Iran war.
The Golden Dome dispute
Michael said the conflict sharpened over debates about AI’s role in Trump’s Golden Dome missile defense program, which aims to place U.S. weapons in space. He offered a hypothetical: if a Chinese hypersonic missile were inbound with only 90 seconds to respond, a human anti-missile operator “may not be able to discriminate with their own eyes what they’re going after,” but an autonomous counterattack in space would be low risk “because it’s in space and you’re just trying to hit something that’s trying to get you.”
“This is part of the debate I had with Anthropic, which is we need AI for things like Golden Dome,” Michael said.
He also described a second scenario: a military base with soldiers asleep, protected by a laser system that autonomously shoots down drones. The implication, he said, was that objections to such uses were irrational given the military’s need to compete with rivals including China.
Anthropic said it sought to limit Claude to two high-level restrictions: no mass surveillance of Americans and no use in fully autonomous weapons. The company argued that today’s leading AI systems “are simply not reliable enough to power fully autonomous weapons.”
The failed negotiations
Michael said Anthropic’s willingness to grant individual scenario exceptions was not sufficient for military planning. Over three months of talks, he said, he walked Amodei through hypotheticals — hypersonic intercepts, drone swarms — and received carve-outs for each.
“I was like, exceptions doesn’t work. I can’t predict for the next 20 years what (are) all the things we might use AI for,” Michael said.
“I need to have the terms of service be rational relative to our mission set,” he added.
The Pentagon ultimately demanded that Anthropic and other AI companies agree to “all lawful use” of their technology. Google, OpenAI, and Elon Musk’s xAI agreed to those terms, Michael said, though some still needed to prepare infrastructure for classified military work. Anthropic did not.
Michael also said the company objected to the Pentagon’s interest in bulk-collecting public information on people using Claude — what he characterized as a second sticking point in the “interminable” negotiations.
Competing accounts
In response to Michael’s podcast remarks, Anthropic pointed to a statement by Amodei: “Anthropic understands that the Department of War, not private companies, makes military decisions. We have never raised objections to particular military operations nor attempted to limit use of our technology in an ad hoc manner.”
Anthropic disputed parts of Michael’s account of the talks and said the restrictions it sought were narrow and not based on existing uses of Claude. The next stage of the dispute, the company said, would likely unfold in court.
Context and background
Michael made the remarks on the “All-In” podcast alongside Silicon Valley venture capitalists Jason Calacanis, David Friedberg, and Chamath Palihapitiya. A fourth regular co-host, former PayPal executive David Sacks — now Trump’s AI czar — did not appear in the episode but has publicly criticized Anthropic, including over its hiring of former Biden administration officials after Trump returned to the White House last year.
Michael said he was sworn in last May as defense undersecretary for research and engineering and took over the military’s AI portfolio in August, at which point he began scrutinizing Anthropic’s contracts, some of which dated from the Biden administration.
As talks were collapsing last week, Michael had attacked Amodei on social media, saying he “has a God-complex” and “wants nothing more than to try to personally control” the military. In the podcast, he framed the dispute more broadly as part of a deliberate military shift toward giving greater autonomy to swarms of armed drones, underwater vehicles, and other machines.