Everyone Is Wrong About China and AI Safety — Sebastian Mallaby Explains

Sebastian Mallaby, a Council on Foreign Relations economist, recounts his March trip to China and challenges prevailing US assumptions about Chinese engagement on artificial intelligence safety. Speaking before his book's US publication, Mallaby met with AI leaders across four Chinese cities at major companies including Huawei, Hikvision, and Ant Group. Contrary to what Biden administration AI policy officials had told him, Mallaby found Chinese technologists and academics unprompted raised AI safety concerns, suggesting an opening for US-China dialogue on preventing dangerous AI proliferation. Mallaby draws parallels to Cold War nuclear non-proliferation efforts, arguing that while the US and China may achieve rough AI parity through mutual deterrence, both nations share interest in preventing criminals, terrorists, and rogue states from accessing powerful open-weight AI models. He advocates for an International Atomic Energy Agency-style framework for AI governance. Notably, Mallaby reverses his 2022 position supporting chip export controls to China, acknowledging the policy has failed to create the anticipated US advantage, with America maintaining only an 8-month lead in frontier models that may disappear when application deployment is factored in. He now argues collaboration with China on AI safety should take priority over maintaining export restrictions, even if that requires loosening controls. The discussion challenges core assumptions underlying current US technology policy toward China and suggests the national security establishment may be missing opportunities for productive engagement on shared risks.

• Sebastian Mallaby found Chinese AI leaders unexpectedly engaged with safety concerns during March trip, contradicting Biden administration claims China refuses to discuss AI risks.
• Mallaby now argues US chip export controls to China have failed after 3.5 years, delivering only an 8-month advantage in frontier AI models that may be nonexistent in applications.
• Council on Foreign Relations economist advocates prioritizing US-China collaboration on AI safety over maintaining chip export restrictions, reversing his 2022 position supporting controls.
• Mallaby proposes Cold War-style non-proliferation framework for AI, comparing potential agreement to 1968 Nuclear Non-Proliferation Treaty negotiations with Soviet Union.
• Both US and China share interest in preventing criminals, terrorists, and rogue states from accessing dangerous open-weight AI models, Mallaby argues.
• Chinese view technology positively as driver of 25-year growth story, contrasting with Western ambivalence rooted in atomic bomb and Cuban missile crisis experience.
• Current administration officials continue to claim China is impossible to negotiate with on AI safety despite evidence of Chinese engagement on the topic.