Defense Secretary Pete Hegseth has given Anthropic, a leading artificial intelligence company, until Friday to agree to broad military access for its AI models, a move that could fundamentally alter the company’s trajectory and set a precedent for the role of AI in national security.
Hegseth is demanding that Anthropic remove stipulations in its contracts that prohibit the utilize of its AI technology for mass surveillance or in the development of lethal autonomous weapons systems. He wants the company to agree that its technology can be used for “any lawful purpose” determined by the Department of War, according to sources familiar with the negotiations.
The ultimatum, first reported by CNBC and the Associated Press, has placed Anthropic in a precarious position. Failure to comply could result in the cancellation of its existing $200 million contract with the Pentagon and a designation as a “supply chain risk,” effectively barring the company from doing business with the U.S. Military and potentially crippling its growth. The company is currently valued at $380 billion and is considering an initial public offering as soon as next year.
A meeting between Hegseth and Anthropic CEO Dario Amodei on Tuesday in Washington, D.C., failed to bridge the gap, with Hegseth reiterating his demands, sources said.
The dispute comes amid a broader push by the Trump administration to assert control over the development and deployment of AI, and has been fueled by criticism from within the administration. AI czar David Sacks has publicly labeled Anthropic’s approach as “woke AI” and accused the company of using safety concerns to stifle competition and secure a regulatory advantage. Sacks has argued that Anthropic is deliberately exaggerating the risks of AI to justify regulations that only it can easily meet.
Anthropic has rejected these claims, with Amodei insisting the company shares many policy goals with the administration, including maintaining U.S. Leadership in AI development. Although, the administration appears determined to press its advantage.
Other AI companies, including OpenAI and Google, have reportedly not imposed similar restrictions on military use of their technologies. In December, the Department of War announced it would deploy Google’s Gemini AI models for a number of applications, following a staff revolt at Google in 2018 over its involvement in “Project Maven,” an AI-powered aerial surveillance program.
According to Owen Daniels, associate director of analysis at the Center for Security and Emerging Technology (CSET) at Georgetown University, “Anthropic’s peers, including Meta, Google and xAI, have been willing to comply with the department’s policy on using models for all lawful applications. So the company’s bargaining power here is limited, and it risks losing influence in the department’s push to adopt AI.”
Anthropic’s stance reflects a deeper internal debate about the ethical implications of AI and the company’s founding principles. The company was established by researchers who left OpenAI over concerns that commercial pressures were overshadowing safety considerations. More recently, Anthropic has publicly committed to responsible AI development, refusing to incorporate advertising into its Claude products or create AI companions designed for romantic interactions.
Hegseth has also indicated that the Pentagon could invoke the Defense Production Act of 1950, originally designed to mobilize civilian manufacturing during wartime, to compel Anthropic to provide unrestricted access to its Claude model. The act was used during the COVID-19 pandemic to boost production of medical supplies and has been invoked several times since by the Biden administration, even outside of declared emergencies.
Legal experts are divided on the legality and wisdom of Hegseth’s approach. Katie Sweeten, a former Department of Justice liaison to the Department of Defense, questioned the logic of potentially utilizing a “supply chain risk” – a designation typically reserved for foreign adversaries – while simultaneously seeking to leverage the company’s technology. Dean Ball, a former AI policy advisor to the Trump Administration, described the Pentagon’s position as “incoherent” in a post on X, questioning the rationale of labeling a company a risk while also considering forcing its compliance.
Alan Rozenshtein, an associate professor of law at the University of Minnesota and a fellow at Brookings, argued that Congress, rather than the Pentagon, should establish the rules governing military AI deployment. “The terms governing how the military uses the most transformative technology of the century are being set through bilateral haggling between a defense secretary and a startup CEO, with no democratic input and no durable constraints,” he wrote.
As of Wednesday, Anthropic had not publicly signaled any intention to alter its position. Helen Toner, interim executive director of Georgetown’s CSET and a former OpenAI board member, suggested the Pentagon may underestimate Anthropic’s reluctance to concede, as doing so could set a problematic precedent for future iterations of its AI models. Anthropic researchers have expressed concerns about how each new version of Claude learns from data, including commentary about the model itself.
Anthropic has demonstrated a willingness to compromise when facing existential threats. In June 2025, the company reached a $1.5 billion settlement in a copyright dispute over its use of copyrighted books to train its AI models. More recently, Anthropic updated its Responsible Scaling Policy, dropping a previous commitment to only train AI models with guaranteed safety controls, opting instead to match the safety efforts of its competitors.
