From TIME · Julian Gewirtz · · 2m
How to make the U.S.-China AI race less dangerous
Instead of sweeping restrictions, the U.S. and China should focus on practical safety measures and crisis communication to prevent AI incidents from spiraling into conflict.
In brief
Instead of sweeping restrictions, the U.S. and China should focus on practical safety measures and crisis communication to prevent AI incidents from spiraling into conflict. The U.S. and China don't need to trust each other to cooperate on AI safety. They just need to recognize that neither benefits if the other loses…
The timing is terrible
Treasury Secretary Scott Bessent is meeting Chinese Vice Premier He Lifeng to discuss AI, but Trump has called AI warnings a hoax. China accuses the U.S. of using safety calls to slow their progress.
Neither side wants to slow down
The U.S. and China don't trust each other and both want to win the AI race. But ignoring calls for dialogue would be a mistake as AI systems advance rapidly.
Avoid the two extremes
Setting talks too low produces meaningless photo ops. Setting them too high chases unverifiable restrictions that will fail. The goal is something in between that actually works.
Focus on AI safety in parallel
Both countries should pursue technical exchanges on testing AI models for dangerous capabilities like bioweapon development or loss of human control. This improves safety without requiring trust.
Self-interest drives cooperation
Neither country benefits if the other loses control of a powerful AI system. Technical exchanges on testing and safeguards serve both nations' interests without surrendering competitive advantage.
Create a crisis hotline
The U.S. and China need a protocol to communicate about AI incidents before they escalate. What if an AI model escapes control and attacks the other country's networks or military?
Write it down, don't call
Initial incident reports should be transmitted in writing, not by phone. This gives Chinese committees time to analyze and respond without individual officials feeling pressured to improvise.
Keep chip controls separate
If China tries to tie AI safety progress to relief from U.S. export controls on advanced chips, Washington should refuse to link the two issues. AI safety talks need their own track.
Start small, build momentum
Neither country is ready for sweeping restrictions. A modest start on safety practices could create channels and expertise that become more valuable as AI capabilities grow and concerns mount.
They don't need to agree on winning
The U.S. and China may never agree on who should lead in AI. But they can agree that neither can afford to lose control of these systems.
The short version
The U.S. and China don't need to trust each other to cooperate on AI safety. They just need to recognize that neither benefits if the other loses control of a powerful system.
