Melanie Sisson of the Brookings Institution and Tianjiao Jiang of Fudan University are urging immediate action to keep artificial intelligence out of the nuclear launch loop. Their joint proposal, released ahead of the scheduled September 24 meeting between Donald Trump and Xi Jinping, outlines strict boundaries for autonomous systems. The core demand is simple but critical: no AI should ever have the authority to initiate a nuclear strike or target command-and-control infrastructure independently. Humans must remain the sole decision-makers when it comes to strategic weapons.
The plan goes beyond just banning autonomous launches. It calls for both nations to agree on a shared definition of “human control” to prevent semantic loopholes from undermining safety. Another key recommendation involves maintaining human oversight over AI-driven cyberattacks on strategic infrastructure. This builds on the November 2024 agreement between President Biden and Xi, aiming to solidify those earlier diplomatic gestures into concrete operational rules. Without clear definitions, the risk of misinterpretation grows as algorithms become more complex and faster than human reaction times.
Jiang also suggests establishing a dedicated hotline for AI-related incidents. The logic is straightforward: if a defensive AI system reacts automatically to suspicious digital activity, the other side might mistake it for an offensive strike. A direct communication channel would allow governments to clarify accidents before they spiral into conflict. However, skepticism remains high. Carla Freeman from Johns Hopkins University points out that diplomatic hotlines are only useful if someone answers. During the 2023 spy balloon crisis, China did not pick up US calls, raising doubts about the reliability of such mechanisms in a true emergency.
While the National Security Commission on Artificial Intelligence flagged this issue back in 2021, binding international mechanisms still do not exist. The gap between expert consensus and political reality is widening as AI capabilities accelerate. These proposals offer a technical roadmap, but their success depends entirely on political will. With tensions high and trust low, the question isn’t whether the technology exists to automate nuclear decisions, but whether rival powers can agree to keep their fingers off the trigger.
