Policy experts from the United States and China have proposed shared rules that would prevent AI systems from independently launching nuclear weapons or attacking nuclear command infrastructure. Melanie Sisson of the Brookings Institution and Tianjiao Jiang of Fudan University published the recommendations before a planned September 24 meeting between Presidents Donald Trump and Xi Jinping.
Their red lines call for humans to retain sole authority over nuclear deployment and AI-enabled cyberattacks on strategic systems. The two countries would also agree on a common definition of meaningful human control, reducing the chance that similar words conceal different operational practices. The proposal builds on a 2024 agreement between Joe Biden and Xi about maintaining human control over nuclear weapons.
Jiang additionally suggests a direct hotline for AI incidents. If an automated defensive system reacts to suspicious activity, one side could otherwise misread the response as a deliberate attack and escalate. A hotline would offer a way to clarify accidents. No binding mechanism currently enforces these ideas, and communication channels are not guaranteed to work: China did not answer US calls during the 2023 spy-balloon crisis. The recommendations therefore identify useful boundaries, but verification procedures and reliable crisis communication would determine whether they reduce risk in practice.