A senior Pentagon official told Anthropic: dispute with government The debate over the use of the company’s artificial intelligence technology in fully autonomous weapons emerged after President Donald Trump’s debate over how AI will be used in the future. Golden Dome Missile Defense Planaimed at deploying U.S. weapons into space.
Under Secretary of Defense Emile Michael, the Pentagon’s chief technology officer, said he has come to believe that the company’s ethical restrictions on the use of its chatbot Claude are an unreasonable hurdle as the U.S. military seeks to give more autonomy to crowds. armed droneunderwater vehicles and other machinery were developed to compete with rivals like China that could do the same.
“I need a reliable, stable partner who can give me something and help me with self-driving because one day it’s going to be a reality and we’re starting to see early versions of that,” Michael said on Friday’s podcast. “We need someone who won’t quit midway through.”
Comments came after the Pentagon officially designated San Francisco-based Anthropic faces the risk of disrupting supply chain defense operations using rules meant to prevent foreign adversaries from damaging national security systems.
Anthropic has vowed to sue over the designation, which affects its business partnerships with other military contractors.
President Trump also ordered federal agencies to immediately stop using Claude, but the Republican president gave the Pentagon six months to phase out products that are deeply embedded in sensitive military systems, including those used in military systems. iran war.
Anthropic said it was simply trying to limit its technology from being used for two high-level applications: mass surveillance of Americans or fully autonomous weapons.
Michael, a former Uber executive, revealed aspects of his months-long talks with Anthropic CEO Dario Amodei during a lengthy conversation with his co-hosts on the All In podcast, Silicon Valley venture capitalists Jason Calacanis, David Friedberg, and Chamath Palihapitiya.
The fourth co-host, former PayPal executive David Sachs, who is now President Trump’s AI czar and was not present, has been a vocal critic of Anthropic, including for hiring a former Biden administration official shortly after Trump returned to the White House last year.
After negotiations stalled last week, Michael blasted Amodei on social media, saying Amodei “has a God complex” and wants “nothing but personal control” over the military. But in the podcast, he framed the conflict as part of a broader military shift toward using AI.
Michael said the military is developing procedures that allow for different levels of autonomy in warfare, depending on the risks posed.
“This is part of the discussion I had with Anthropic, that things like Golden Dome require AI,” Michael said, sharing a hypothetical scenario in which the U.S. had only 90 seconds to respond to a Chinese hypersonic missile.
While human anti-missile operators “might not be able to discern with their own eyes what they’re aiming at,” an autonomous counterattack would be less risky “because you’re just in space trying to hit whatever’s trying to get you.”
In another scenario, he said, “If you have a military base, a bunch of soldiers sleeping, and a laser that can autonomously destroy drones, who can object?”
In response to comments on the podcast, Anthropic pointed to Amodei’s previous statement, saying, “Anthropic understands that military decisions are made by the Department of the Army, not private companies. We have never objected to specific military operations or sought to limit the use of our technology in an ad hoc manner.”
Michael, the undersecretary of defense for research and engineering, was sworn in last May and said he took over the military’s “AI portfolio” in August. He said that’s when he began reviewing Anthropic’s contracts, some of which date from the Democratic administration of President Joe Biden. Michael said he questioned Anthropic’s terms of service, which he believed were too restrictive.
“Terms of service need to be reasonable in light of our mission,” he said. “So we started negotiating. It took three months, but we had to present them with certain scenarios, like this Chinese hypersonic missile example, and they were like, ‘Okay, we’ll make an exception for that.'” So what about this swarm of drones? “I’ll make an exception to that.” And I thought, exceptions don’t work. There is no way to predict what we might be using AI for over the next 20 years. ”
Michael said the Pentagon has begun insisting that Anthropic and other AI companies be allowed “all lawful uses” of their technology.
humans resisted the changeclaims that today’s leading AI systems are “just not reliable enough to power fully autonomous weapons.”
Competitors Google, OpenAI and Elon Musk’s xAI have agreed to the Pentagon’s terms, but some companies still need to prepare their infrastructure for sensitive military operations, Michael said. Another problem with Anthropic was that it did not allow mass surveillance of Americans.
“They didn’t want us to use AI systems to collect a lot of public information about people,” Michael said, describing the negotiations as “endless.”
Antropic disputed some of the content of Michael’s talks, stressing that the protections sought were narrow in scope and not based on Claude’s existing use. The next step in the dispute is likely to be in court.
