US and China experts push for shared rules banning AI control over nuclear weapons

A coalition of weapons policy and AI experts from the United States and China is pushing for binding international agreements to prevent artificial intelligence from making autonomous decisions about nuclear weapons deployment. The effort represents a rare point of collaboration between the two superpowers on an issue where miscalculation could trigger catastrophic consequences.

The experts argue that allowing AI systems to independently launch nuclear weapons creates unacceptable risks. Current nuclear command structures rely on human decision-makers at critical junctures, but as military forces increasingly integrate autonomous systems into defense infrastructure, the pressure to automate launch decisions grows. Faster automated responses could theoretically reduce reaction time in a crisis, but they also eliminate human judgment from the most consequential military decision any nation can make.

The proposal calls for explicit prohibitions on what researchers term "human-out-of-the-loop" nuclear decision-making. This means AI could assist in analysis, targeting, and threat assessment, but final authorization for any nuclear weapon use must remain with human commanders. The distinction matters because many modern AI systems already influence military planning. The debate centers on whether to preserve human veto power over AI recommendations.

American experts involved in the initiative include specialists from defense think tanks and arms control organizations who have tracked the intersection of artificial intelligence and nuclear strategy for years. Chinese participants bring perspectives from their own defense modernization efforts. Both sides recognize that autonomous nuclear decision-making by one nation could trigger destabilizing arms races where competitors rush to deploy their own autonomous systems out of fear of being disadvantaged.

The timing reflects growing anxiety about AI capabilities. Large language models and autonomous systems have advanced faster than international policy frameworks can accommodate. Nuclear weapons remain the only military domain where two nations can still destroy each other within hours. Adding autonomous AI to that equation amplifies risks of accidents, misinterpretation, or system failures that could initiate war.

Establishing shared rules faces substantial obstacles. International treaties require ratification and verification mechanisms. Nations historically resist constraints on their own weapons while suspecting others of cheating. The US and China have limited history of successful nuclear cooperation agreements outside basic communication channels. Trust between the superpowers remains low across most domains.

The experts frame their proposal not as arms control advocacy but as technical risk management. They argue that preserving human control protects both nations' interests. A false alarm or system malfunction could trigger escalation neither side intended. Keeping humans in decision loops provides circuit-breakers that prevent technical failures from causing war.

Concrete proposals include establishing communication protocols about AI use in nuclear command systems, creating transparency measures where feasible, and potentially negotiating bilateral agreements that later become multilateral standards. Some experts suggest starting with confidence-building measures, like declarations about which nuclear systems will never use autonomous launch authority.

Success remains uncertain. But the convergence of American and Chinese experts on this issue signals recognition that some AI risks transcend national competition. Both nations face identical dangers from autonomous nuclear systems, regardless of which side deploys them first. That shared vulnerability may provide sufficient common ground to move from expert consensus toward binding agreements that constrain AI's role in nuclear decision-making.