Experts from the US and China proposed rules against autonomous AI control over nuclear weapons
Experts from the US (Melanie Sisson, Brookings) and China (Tianjiao Jiang, Fudan University) proposed rules ahead of the Trump-Xi meeting on 24 September prohibiting AI from autonomously launching nuclear weapons or attacking nuclear command systems. Binding mechanisms are still lacking.
Melanie Sisson from Brookings Institution and Tianjiao Jiang from Fudan University have published a concrete proposal for rules intended to prevent AI systems from autonomously deciding on the use of nuclear weapons. The proposal was published ahead of the planned Trump-Xi meeting on 24 September 2026.
The proposal sets out several red lines: no AI system should be allowed to launch nuclear weapons on its own or attack nuclear command systems. Humans must have exclusive control over AI-driven cyberattacks on strategic infrastructure. The two countries should also agree on a shared definition of “human control”. Jiang additionally proposes establishing a hotline for reporting AI incidents – if a defensive AI system automatically responds to suspicious activity, the other side could interpret it as an attack, and a direct line would allow the situation to be clarified in time as a misunderstanding.
The proposal builds on an agreement reached by Biden and Xi in November 2024. The US National Security Commission on Artificial Intelligence (NSCAI) issued a similar warning about maintaining human control over nuclear weapons back in 2021, but no binding mechanisms exist yet. Carla Freeman from Johns Hopkins University questions whether a hotline would work at all – during the spy balloon crisis in 2023, China did not respond to calls from the US.
Why it matters
This is a proposal by academics, not an approved policy – it remains to be decided whether and how the Trump-Xi talks might take it into account. The proposal shows where the debate on limiting the role of AI in military decision-making with the highest stakes is heading, but the sources also show that similar recommendations have existed since 2021 without a binding outcome and that even the proposed communication channel (hotline) has real historical limitations, according to skeptics.
Check the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.