Jack Paul 
Bill Gates is calling for a new international system to handle the riskiest uses of artificial intelligence. His argument: individual governments and AI companies are no longer able to manage increasingly complex models on their own. When he travels to China in November, he hopes to meet President Xi Jinping and talk about AI safety. According to Reuters, Gates thinks China might accept limits on releasing especially dangerous AI models if the United States takes the lead.
This goes beyond usual tech regulation. Gates wants governments to watch whether a sophisticated model can, say, create molecules or offer features that make biological attacks easier. He also sees cybersecurity as a growing issue, because AI systems are getting better at finding and exploiting software flaws. Recent events seem to support that. OpenAI announced this month that assessments of one future model showed enough progress in cybersecurity and agentic coding that the company could no longer exclude “critical” cyber capabilities from its safety categories. That kind of result makes the conversation feel less hypothetical.
Learning from aviation and nuclear oversight
Gates thinks the same methods used for other technologies with cross-border risks could work for AI. One idea is an international regulatory body that borrows from aviation safety and nuclear inspection. The point would be monitoring high-risk capabilities, not stopping ordinary AI development. That distinction matters. Nobody wants a heavy system that slows down useful work. But there is also little room for vague promises when the stakes are this high.
His concerns are not limited to security. Gates has repeatedly warned that AI could replace white-collar and blue-collar jobs faster than earlier technological shifts. He has floated ideas like stronger economic safety nets, taxes on labor replaced by AI, and even reserving certain jobs for people. These are harder to fit into an international agreement, but they are part of the same problem.
China and the United States need each other here
Any serious oversight system would require cooperation between geopolitical rivals. China has already said it supports a global plan for AI governance, including risk-based testing, stricter cybersecurity rules, and joint action against malicious AI use. The catch is whether the U.S. and China can trust each other enough to make common rules stick. Gates sees that coordination as essential. If countries building the most powerful models do not agree on shared limits, dangerous AI work could simply move to another jurisdiction. One nation’s careful rules would just push the problem elsewhere.
The proposal is ambitious, and arguably a bit optimistic. But it reflects a real concern: AI is moving faster than governance. Whether Gates can help bring Washington and Beijing to the same table remains an open question. Still, he seems willing to try.