Defense Secretary Pete Hegseth met with Anthropic CEO Dario Amodei on Tuesday and delivered an ultimatum: agree to let the Pentagon use its AI model without restrictions by 5:01 p.m. Friday, or face consequences that could cost the company its government contracts and potentially its ability to work with any firm that does business with the Defense Department.

The meeting was serious in substance if not in tone. People familiar with the discussions described it as cordial throughout, with no raised voices and Hegseth praising the quality of Anthropic's technology. What the two sides could not bridge was the central question that has been building for months: whether a private company can place limits on how the military uses its products, even when those products are integrated into classified systems and active operations.

Amodei did not move on the two areas he has publicly defined as non-negotiable — the use of Claude to power autonomous weapons that fire without human oversight, and the use of Claude for mass surveillance of American citizens. A source familiar with the discussions said Anthropic has no intention of dropping those restrictions.

How the Dispute Reached This Point

The immediate trigger was January's raid to capture Venezuelan President Nicolás Maduro. Claude, deployed through Anthropic's partnership with defense technology firm Palantir, was used in preparation for the operation, during which scores of Maduro's security guards and Venezuelan service members were killed.

After the raid, a senior defense official said, an Anthropic executive contacted a counterpart at Palantir asking whether Claude had been used in the operation — a question that Palantir relayed to the Pentagon, along with the interpretation that Anthropic might have had reservations about how its technology was applied. That account prompted Pentagon leadership to question whether Anthropic could be fully relied upon as a defense partner.

Anthropic disputed this characterization directly. Amodei told Hegseth during the Tuesday meeting that neither he nor anyone at the company raised concerns about the Maduro raid with Palantir or with the Pentagon, and that any communication with Palantir had been limited to routine technical discussions. He also said the company's existing restrictions have never posed a problem for any Defense Department operation currently underway.

The deeper friction predates the Venezuela raid. In January, Hegseth issued a directive ordering the Pentagon to accelerate AI adoption, instructing the department to move "from campaign planning to kill chain execution" and to "approach risk tradeoffs, equities, and other subjective questions as if we were at war." Days after that directive, Amodei published an essay warning of the dangers of AI-powered autonomous weapons and mass surveillance systems, writing that such tools in the hands of governments with limited accountability could be used to "detect pockets of disloyalty forming and stamp them out before they grow." He did not name any specific government.

What the Pentagon Has Threatened

Hegseth outlined two options if Anthropic does not agree to the Pentagon's terms by Friday.

The first is invoking the Defense Production Act, which gives the president authority to compel private companies to prioritize and fulfill contracts required for national defense. It was used during the COVID-19 pandemic to increase production of vaccines and ventilators. Whether it could be used to force a software company to remove safety restrictions from a proprietary AI model is a question legal experts say has no settled answer.

"I'm not sure that's how that part of DPA has tended to be used, or has ever been used," said Jerry McGinn, a director at the Center for Strategic and International Studies for industrial base issues.

Franklin Turner, a government contracts lawyer at McCarter & English, said the situation was "unprecedented and will almost certainly trigger a raft of downstream litigation" if the administration takes adverse action against Anthropic.

The second option is designating Anthropic a supply chain risk — a designation that has historically been applied to Chinese and Russian companies. That classification would require every Pentagon vendor and contractor to certify they are not using Anthropic's models in their workflows, which would effectively lock the company out of the broader defense industrial ecosystem regardless of whether its direct contract was terminated.

The Pentagon's Leverage Problem

The dispute puts the Defense Department in an awkward position it has acknowledged internally. Claude is currently the only AI model approved for use on classified military networks, where it handles a range of functions from intelligence analysis and cybersecurity work to more routine administrative tasks. Elon Musk's Grok recently received clearance for classified use, and the Pentagon has been accelerating conversations with OpenAI and Google about moving their models — currently authorized only for unclassified systems — into classified settings.

"The only reason we're still talking to these people is we need them and we need them now," a senior Defense official said of Anthropic ahead of the Tuesday meeting. "The problem for these guys is they are that good."

One person familiar with the discussions said Claude currently holds an advantage over its competitors in several categories relevant to military applications, including offensive cyber capabilities. Gemini, Google's model, has been identified as the most likely replacement if a deal is reached, though that would require Google to agree to the same "all lawful purposes" terms Anthropic has declined. OpenAI and xAI have already agreed to those terms for unclassified use and are working toward classified agreements.

What Anthropic Has Said

Anthropic has maintained a consistently measured public posture while standing firm on the substance.

"We continued good-faith conversations about our usage policy to ensure Anthropic can continue to support the government's national security mission in line with what our models can reliably and responsibly do," a company spokesperson said after the Tuesday meeting. "During the conversation, Dario expressed appreciation for the Department's work and thanked the Secretary for his service."

The company holds a two-year, $200 million Pentagon contract secured last year, covering Claude Gov — a version of Claude built for national security work — along with enterprise products used more broadly within the military. Losing that contract would be a significant financial blow, but people familiar with Amodei's position say the company is prepared to walk away rather than remove restrictions on autonomous weapons and domestic surveillance.