Pentagon Issues Friday Deadline to Anthropic Over Military AI Restrictions

The ultimatum delivered by Defense Secretary Pete Hegseth to Anthropic CEO Dario Amodei marks a watershed moment in the relationship between the U.S. government and the artificial intelligence industry. By threatening to invoke the Defense Production Act (DPA) to compel Anthropic to lift restrictions on its technology, the Pentagon is signaling that it views high-level AI models not merely as commercial products, but as critical national security infrastructure. This move places Anthropic, a company founded on the principles of AI safety and ethical alignment, in a precarious position between its corporate mission and the demands of the state.

The core of the dispute lies in Anthropic’s refusal to allow its Claude model to be used for two specific purposes: fully autonomous military targeting and domestic surveillance of U.S. citizens. Amodei has been vocal about these risks, recently writing that a powerful AI could be used to gauge public sentiment, detect pockets of disloyalty forming, and stamp them out. This stance contrasts sharply with the Pentagon’s current leadership, which has vowed to purge what it terms woke culture from the military and views any restriction on technological capabilities as a hindrance to national defense.

“

The ultimatum delivered by Defense Secretary Pete Hegseth to Anthropic CEO Dario Amodei marks a watershed moment in the relationship between the U.S.

Anthropic is reportedly the last major AI lab to resist joining a new internal U.S. military network, a move that its peers—including OpenAI, Google, and Elon Musk’s xAI—have largely accepted or are already facilitating. This isolation makes Anthropic a prime target for the administration’s supply chain risk designation, a label that could effectively blacklist the company from future federal contracts and potentially chill its relationships with private sector partners who fear secondary regulatory pressure.

The potential invocation of the Defense Production Act for software and AI models is a significant escalation of executive power. Historically used for physical goods like steel during wartime or vaccines during a pandemic, applying the DPA to a Large Language Model suggests the government believes it has the right to commandeer the weights or the operational logic of an AI system. If the Pentagon follows through, it could set a precedent where no AI developer can legally prevent the military from repurposing their technology for lethal or surveillance applications, regardless of the company's terms of service.