Pentagon Issues Friday Ultimatum to Anthropic Over Military AI Guardrails

The confrontation between the Department of Defense and Anthropic marks a definitive end to the 'honeymoon phase' of voluntary AI safety commitments in the United States. Defense Secretary Pete Hegseth’s ultimatum to CEO Dario Amodei—to either permit the full military application of Claude AI or face contract termination—strikes at the heart of Anthropic’s corporate identity. Since its inception as a spin-off from OpenAI, Anthropic has positioned itself as the 'safety-first' AI lab, prioritizing 'Constitutional AI' and rigorous guardrails. However, the Department of Defense (DoD) now views these internal ethical constraints as a strategic liability that could impede American technical superiority on the battlefield.

At the center of the dispute are two specific prohibitions maintained by Anthropic: the use of its models for fully autonomous lethal targeting and the mass surveillance of U.S. citizens. While Amodei argues these safeguards are essential to prevent catastrophic misuse, the Pentagon views them as unacceptable limitations on a $200 million contract. The threat to invoke the Defense Production Act (DPA) is particularly significant. Historically used to compel the production of physical goods like steel or medical supplies, the application of the DPA to generative AI software would set a radical precedent, effectively treating private-sector code as a state-controlled resource during times of perceived national necessity.

“

While Amodei argues these safeguards are essential to prevent catastrophic misuse, the Pentagon views them as unacceptable limitations on a $200 million contract.

This escalation also highlights a widening rift in the competitive landscape of 'Defense Tech.' While Anthropic was the first to achieve the high-level security clearances required for the Pentagon’s classified networks via its Claude Gov product, its rivals have been more accommodating to the military's shifting requirements. Google and Elon Musk’s xAI have reportedly signaled a greater willingness to integrate their models into kinetic and surveillance operations, earning public praise from Hegseth. For Anthropic, which recently reached a staggering $380 billion valuation, losing the Pentagon’s business is not just a financial blow; it risks a 'supply-chain risk' designation that could freeze the company out of the broader federal market, which is increasingly the primary driver of large-scale AI infrastructure spending.