Pentagon Clashes with Anthropic Over AI Autonomy for 'Golden Dome' Defense

The revelation by U.S. Defense Undersecretary Emil Michael regarding the Pentagon’s fallout with Anthropic marks a watershed moment in the relationship between the American defense establishment and the generative AI industry. At the heart of the dispute is the Golden Dome missile defense program, a cornerstone of President Donald Trump’s national security strategy that envisions a space-based shield capable of intercepting threats at hypersonic speeds. Michael, the Pentagon’s chief technology officer and a former Uber executive, argued that the ethical restrictions Anthropic placed on its Claude model—specifically prohibiting its use in fully autonomous lethal systems—constituted an irrational obstacle to national security.

The friction highlights a fundamental divergence in philosophy between Silicon Valley’s AI safety movement and the Department of Defense's operational requirements. Anthropic, founded on the principle of AI safety, has long maintained that its technology should not be used for mass surveillance or fully autonomous weaponry. However, the Pentagon views autonomy not as a choice, but as a technical necessity in the modern theater of war. As Michael noted during his appearance on the All-In podcast, the speed of modern missile threats and the complexity of managing drone swarms or underwater vehicles require machines that can make split-second decisions without a human in the loop for every action. The Pentagon’s fear is that if the U.S. adheres to strict ethical constraints while rivals like China aggressively pursue autonomous capabilities, the American military will be left with a decisive disadvantage.

“

Defense Undersecretary Emil Michael regarding the Pentagon’s fallout with Anthropic marks a watershed moment in the relationship between the American defense establishment and the generative AI industry.

The consequences for Anthropic have been swift and severe. The Pentagon’s designation of the San Francisco-based firm as a supply chain risk is a potent regulatory move that effectively blacklists the company from defense contracts and complicates its partnerships with major military integrators. This designation is typically reserved for companies with ties to foreign adversaries, making its application to a prominent American AI lab a significant escalation. Furthermore, President Trump’s order to phase out Claude from all federal agencies within six months—despite its deep integration into classified systems used in active conflicts like the Iran war—signals a total breakdown in trust between the administration and the AI startup.

This rift creates a vacuum that other AI developers are already moving to fill. While Anthropic has vowed to sue the government over the supply chain designation, the Pentagon is actively seeking reliable, steady partners who will not hesitate when asked to develop autonomous lethal systems. This shift favors companies like Palantir, Anduril, and potentially OpenAI, which recently softened its stance on military applications. The message from the Pentagon is clear: in the era of Great Power competition, the safety priorities of Silicon Valley will be secondary to the operational requirements of the Department of Defense.

Looking ahead, the legal battle between Anthropic and the U.S. government will likely set a precedent for how much control private companies can exert over the end-use of their dual-use technologies. If the government successfully argues that national security overrides a company’s ethical terms of service, it could fundamentally change the business model for AI labs seeking federal funding. For the Golden Dome program, the immediate challenge will be the technical migration away from Claude. Replacing a deeply embedded AI model in classified environments is a high-risk endeavor that could delay the deployment of the space-based defense shield, even as the administration pushes for rapid acceleration to counter Chinese and Russian advancements.