AI company Anthropic said it could not accept the Pentagon’s “best and final” offer to resolve a dispute over restrictions the company has in place on how the U.S. military can use its AI models. With just hours left before a Friday deadline to comply with the Pentagon’s demands or face actions that could see Anthropic barred from doing business with any company that also does business with the U.S. military, the dispute turned increasingly ugly.
Pentagon officials have publicly questioned the character of Anthropic CEO Dario Amodei. Meanwhile, employees at competing AI labs have signed open letters supporting Anthropic’s position. OpenAI CEO Sam Altman told his employees in a memo on Thursday, according to reporting from Axios, that OpenAI would push for the same limitations on autonomous weapons and mass surveillance that Anthropic has as it negotiates to extend the use of ChatGPT, currently available to the military for nonsensitive use cases, to more classified domains.
The Anthropic-Pentagon fight is now threatening to spiral into an industrywide rebellion among tech workers at AI companies over how the AI systems they are building are used by the military. On Thursday, more than 100 workers at Google sent a letter to Jeff Dean, the company’s chief scientist, also asking for similar limits on how Google’s Gemini AI models are used by the U.S. military, according to the New York Times.
On Thursday, Amodei published a lengthy statement explaining why the company believes there should be restrictions on the use of his company’s AI technology for autonomous weapons and mass surveillance. These are the two areas where Anthropic currently restricts use of its models by the military, both in its contract terms and through safeguards it has built directly into its Claude models. The Pentagon wants these limitations removed and for Anthropic to agree that the U.S. military can use its models “for any lawful purpose.”
Frontier AI systems are “not reliable enough to power fully autonomous weapons,” and without proper oversight, they “cannot be relied upon to exercise the critical judgment that our highly trained, professional troops exhibit every day,” Amodei wrote in his statement. On surveillance, he argued, powerful AI can now stitch together individually innocuous public data, such as location records, browsing history, and social associations, into a comprehensive portrait of any American citizen’s life at scale.
Emil Michael, the U.S.’s undersecretary of war, called Amodei “a liar” with a “God complex” in response, accusing the CEO of wanting “to personally control the U.S military” in posts on social media platform X. In a separate post, Michael also characterized Anthropic’s Claude Constitution—an internal document outlining the values and principles the company builds into its AI—as a corporate plot to “impose on Americans their corporate laws.”
The Pentagon has demanded Anthropic remove the contract limitations it objects to by 5:01 p.m. Friday or face having its $200 million contract with the U.S. military canceled or, in a more extreme move, be labeled “a supply-chain risk,” which would effectively bar any company doing business with the military from using Anthropic’s technology.
This kind of step is normally reserved for the companies of foreign adversaries such as China’s Huawei or Russian cybersecurity firm Kaspersky.
“Using it against a domestic company for reasons of them not being willing to bend on some principles of this sort is really quite escalatory and unprecedented,” Seán Ó hÉigeartaigh, executive director of Cambridge’s Center for the Study of Existential Risk, told Fortune.
The Department of War has also threatened to invoke the Cold War–era Defense Production Act, using the law to compel Anthropic to hand over an unrestricted version of Claude on the grounds that the government deems it essential to national security. If the Pentagon does go down this route, it will be using powers intended only for emergencies to resolve a contract dispute during peacetime. There is some precedent for this: The Biden administration also invoked the DPA in 2023 to compel frontier AI labs to hand over information about the safety of their AI models. But compelling a company to produce a product, as opposed to simply provide information, comes closer to nationalization of a leading technology company.
“If they are being effectively coerced into allowing their technology to be used in ways that even they themselves say is not reliable in high-stakes life and death situations like on the battlefield,” Ó hÉigeartaigh said, “that sets a very dangerous precedent.”
The Department of War has publicly stated it has no intention of conducting mass surveillance or removing humans from weapons-targeting decisions, but the dispute could rest on how each side is defining “autonomous” or “surveillance” in practice. Representatives for the department did not immediately respond to a request for comment from Fortune.
An Anthropic spokesperson told Fortune that the company was continuing “to engage in good faith” with the Department of War. However, the spokesperson said that contract language received overnight had made “virtually no progress” on the core issues. New language “framed as compromise” was “paired with legalese that would allow those safeguards to be disregarded at will,” they said. Amodei has called the threats from the Department of War “inherently contradictory” as “one labels us a security risk; the other labels Claude as essential to national security.”
Anthropic has won praise from some corners for its willingness to stand firm. Harvard law professor Lawrence Lessig praised the company’s statement as “a beautiful act of integrity and principle” and called it “incredibly rare for our time.”
Rivals OpenAI and xAI have agreed to Pentagon contracts that allow their models to be used for all lawful purposes, with xAI going further by also agreeing to deploy its systems in some classified settings. But more than 330 current employees at rival labs Google DeepMind and OpenAI have also published an open letter in support of Anthropic that urges their own leadership to follow the company’s lead. “They’re trying to divide each company with fear that the other will give in,” the letter read. “That strategy only works if none of us know where the others stand.” The signatories included senior research scientists and both named and anonymous researchers from both companies.
Ó hÉigeartaigh said that the outcome of the dispute could extend well beyond Anthropic itself. “If the Pentagon comes out on top of this,” he said, “it will establish precedents that will not be good for the independence of these companies, or their ability to hold to ethical standards.”
