Anthropic Pentagon AI tensions have surged over the past two weeks, as Anthropic CEO Dario Amodei and Defense Secretary Pete Hegseth publicly spar over how the U.S. military should be allowed to use advanced AI systems.

At the center of the conflict is a fundamental question: who ultimately controls powerful AI models — the companies that build them or the government agencies that want to deploy them?

Anthropic has drawn a firm line. The company refuses to allow its models to be used for AI mass surveillance of Americans or for AI autonomous weapons capable of conducting strikes without human oversight.

Secretary Hegseth, however, argues that the Department of Defense should not be restricted by a vendor’s internal rules. From his perspective, any lawful use of the technology should remain on the table.

What Anthropic Is Concerned About

Anthropic’s leadership has consistently said its AI systems are not yet capable of safely supporting high-stakes military decision-making without strict safeguards.

One major concern involves AI autonomous weapons. While the U.S. military already relies on highly automated systems — some of them lethal — the final decision to use force has historically remained in human hands. However, a 2023 Department of Defense directive allows AI systems to select and engage targets without human intervention, provided certain review standards are met.

That flexibility is precisely what worries Anthropic. If lethal decision-making were increasingly automated, the public might not know until such systems were already operational. And if those systems relied on Anthropic’s models, the use could technically qualify as “lawful.”

The company’s argument is not that such deployments should be permanently banned. Rather, it believes its models are not yet robust enough to handle irreversible battlefield decisions without unacceptable risk.

AI mass surveillance is another flashpoint. Under current U.S. law, surveillance of citizens can occur through collection of communications such as emails and text messages. AI dramatically expands those capabilities by enabling automated pattern detection, entity resolution across datasets, predictive risk scoring, and continuous behavioral analysis at scale.

Anthropic fears that combining powerful generative systems with large-scale data access could supercharge lawful surveillance in ways that outpace democratic oversight.

What the Pentagon Is Demanding

The Pentagon’s stance is straightforward: it should be able to use Anthropic’s technology for any lawful purpose deemed necessary for national defense.

Sean Parnell, the Pentagon’s chief spokesperson, stated on X that the department has no interest in conducting AI mass surveillance domestically or deploying fully autonomous strike systems. Instead, he framed the request as common sense — allowing lawful use without vendor-imposed restrictions.

Hegseth has reportedly given Anthropic until 5:01 p.m. ET on Friday to agree. Otherwise, the company could be designated a supply chain risk, effectively blacklisting it from government contracts. The Pentagon has also floated the possibility of invoking the Defense Production Act to compel compliance.

At times, the debate has also taken on a cultural dimension. In a January speech at SpaceX and xAI offices, Hegseth criticized what he described as “woke AI,” signaling broader ideological friction alongside policy disagreements.

What Happens Next

The stakes in the Anthropic Pentagon AI dispute are significant for both sides.

Sachin Seth, a venture capitalist at Trousdale Ventures who focuses on defense technology, told TechCrunch that being labeled a supply chain risk could be devastating for Anthropic. At the same time, he warned that removing the company from Department of Defense projects could create a national security gap.

According to Seth, replacing Anthropic’s models with alternatives from OpenAI or xAI could take six to twelve months. During that window, the military might operate with second- or third-tier AI capabilities rather than best-in-class systems.

xAI is reportedly preparing to become classified-ready and may be willing to grant the Department broader control over its models. Recent reports suggest OpenAI may maintain safeguards similar to Anthropic’s, though final positions remain fluid.

For now, the Anthropic Pentagon AI standoff underscores a defining tension of the AI era: balancing innovation, national security, and the ethical risks of AI autonomous weapons and AI mass surveillance.