Tensions have escalated between the U.S. Department of Defense and Anthropic, the only artificial intelligence company currently operating a model within classified military systems, after the Pentagon issued a firm deadline demanding compliance with its contractual terms.

According to a senior defense official, Defense Secretary Pete Hegseth warned that if Anthropic does not agree to the department’s conditions by 5:01 p.m. Friday, the administration could invoke the Defense Production Act. Such a move would allow the government to compel access to the company’s technology for national defense purposes and potentially designate Anthropic as a supply chain risk—jeopardizing its existing government contracts.

The dual threats have drawn attention because they appear contradictory. One measure could restrict the government’s use of Anthropic’s products by labeling the company a risk, while the other would force the firm to provide access to its models. Observers say the approach underscores both the Pentagon’s frustration and the strategic importance of advanced AI tools in modern defense operations.

Jessica Tillipman, associate dean at the George Washington University Law School, noted that invoking national security authorities as leverage in a contract dispute raises broader concerns. She suggested that using tools designed for emergency defense mobilization in a commercial disagreement could dilute their long-term credibility.

The dispute intensified following a meeting Tuesday morning at the Pentagon between Secretary Hegseth and Anthropic Chief Executive Dario Amodei. Individuals familiar with the discussion described the conversation as professional in tone. However, when Anthropic declined to accept the department’s proposed terms, the secretary reportedly reiterated the administration’s willingness to escalate.

People on both sides of the negotiations, speaking on condition of anonymity due to the sensitivity of classified systems, described a fundamental disagreement over how artificial intelligence models deployed in military environments should be governed.

Anthropic has argued that its requests are narrowly tailored and reasonable. The company is seeking formal assurances that its AI systems will not be used for mass domestic surveillance or deployed in fully autonomous weapons systems—such as drone operations—without meaningful human oversight. Executives have maintained that such safeguards align with the company’s publicly stated commitment to responsible AI development.

Supporters of Anthropic contend that the firm is facing disproportionate pressure in part because it was the first AI provider to operate inside a classified defense environment. To meet military requirements, the company developed a specialized version of its model known as Claude Gov, adapted for government use and lacking some of the guardrails present in public-facing versions.

Pentagon officials, however, have pushed back on the idea that contractors should dictate operational parameters. They argue that ensuring lawful use of software and weapons systems is the responsibility of the Department of Defense itself. In their view, AI contracts must clearly state that the military may employ the technology for any lawful purpose, without additional constraints imposed by vendors.

While the Defense Production Act grants broad authority to prioritize and compel production in the interest of national defense, it has traditionally been used in industrial or manufacturing contexts. Applying it to a software developer—particularly in a way that might require the company to provide access to its model without standard commercial terms—would be an unusual step.

An Anthropic spokesperson said the company continues to engage in good-faith discussions and remains committed to supporting U.S. national security objectives. At the same time, the spokesperson emphasized that Anthropic must ensure its systems are used in ways consistent with what the technology can responsibly and reliably support.

Defense officials have countered that the disagreement is not about surveillance or autonomous weapons, but about maintaining consistent contractual standards. They say the department’s position is straightforward: artificial intelligence models integrated into classified systems must be available for any application that complies with U.S. law.

Meanwhile, the Pentagon is exploring alternative AI providers. The senior defense official confirmed that the department has reached an agreement with xAI to integrate its Grok model into classified systems. However, significant technical work remains before Grok can be deployed on secure cloud infrastructure and embedded within platforms operated by Palantir Technologies.

Notably, many defense personnel view Anthropic’s Claude model as outperforming Grok in accuracy and analytical reliability, which adds urgency to the ongoing negotiations. The Pentagon is also reportedly close to finalizing an arrangement with Google to deploy its Gemini model within classified environments, though that agreement has not yet been completed.

Another point of tension emerged after Palantir relayed concerns about a reported conversation between one of its employees and an Anthropic counterpart regarding a U.S. military operation last month aimed at capturing Venezuelan President Nicolás Maduro. According to individuals briefed on the meeting, Amodei told defense officials that there had been a misunderstanding and that Anthropic had not initiated contact with either Palantir or the Pentagon concerning the operation.

Amodei also reiterated that his company has never sought to obstruct or interfere with legitimate military missions. Instead, he emphasized that Anthropic’s objective is to establish clear boundaries ensuring that advanced AI systems are deployed in ways that align with both legal standards and the company’s stated ethical commitments.

The standoff illustrates a broader debate unfolding across Washington and Silicon Valley: how to reconcile national security imperatives with emerging norms around responsible AI development. As artificial intelligence becomes increasingly central to defense strategy, the outcome of this dispute could shape not only Anthropic’s future government contracts but also the framework governing how private AI firms collaborate with the U.S. military in the years ahead.