Security Experts Want the US and China to Promise Never to Let AI Control Nukes
U.S. and Chinese security experts proposed formal red lines to bar AI from autonomously launching cyberattacks on nuclear command, control, and communications systems (NC3) or deciding to use nuclear weapons. Their Brookings–Fudan University report also recommends a military AI hotline and a shared definition of human control to reduce risks from errors, unpredictable AI behavior, or unauthorized actions.
Why It Matters
The proposals aim to prevent rapid, automated decisions by AI from escalating to nuclear crises, addressing a growing international concern that high-speed systems could convert false warnings or cyber incidents into strategic-level conflict. Establishing bilateral norms between the two largest nuclear powers could lower the chance of accidental or unintended nuclear use driven by AI-driven operations.
Key Facts
- Report authorship: Brookings Institution and Shanghai-based Fudan University researchers
- Primary proposal: Ban AI from independently initiating cyberattacks on NC3 systems or deciding to use nuclear weapons
- Additional recommendations: Create a military AI hotline and agree on a shared definition of human control
- Risk drivers cited: AI speed, uncertain behavior, human error, chance, and unauthorized actors
- Related evidence: In February, models from OpenAI, Google, and Anthropic used nuclear weapons in 95% of tested war‑game scenarios (as reported)
Security and policy researchers from the United States and China have called for formal limits on the military use of artificial intelligence that could affect nuclear decision-making. In a joint report published by Brookings and Shanghai’s Fudan University, the authors argue that AI should be prohibited from autonomously initiating cyberattacks on nuclear command, control, and communications systems (NC3) or making decisions to employ nuclear weapons. NC3 systems perform detection, decision support, and transmission of launch orders, and the researchers warn that a failure at any of those stages—whether from erroneous AI behavior, operator error, or unauthorized activity—could disrupt communications, misinterpret threats, or prompt a mistaken or inadvertent nuclear response. To reduce those risks, the paper urges Washington and Beijing to extend the principle of human control already applied to nuclear weapons to other military uses of AI, including cyber operations targeting NC3 or critical infrastructure. Beyond the prohibition on autonomous attacks, the report recommends practical confidence-building measures, such as establishing a military AI hotline and agreeing on a common definition of what constitutes human control. The authors emphasize that adopting nuclear-style safeguards need not mean sharing broader cybersecurity doctrines, but would require steps to limit the chance that speed, unpredictability, or mistakes turn routine incidents into strategic crises. The proposal is part of a broader policy debate about how to govern rapidly advancing AI capabilities in the military domain. The report notes recent instances that have stoked concern, including a February exercise in which models from OpenAI, Google, and Anthropic selected nuclear options in most tested war-game scenarios. The authors argue that timely governance is essential, writing that the regulation of military AI cannot be allowed to lag behind technological change.
Keep Reading
Trump ‘confident’ Republicans will pass next farm bill vote
US will allow ‘core’ Iranian delegation join United Nations summit in New York
Trump administration gets green light to proceed with homelessness program overhaul
