US and China Plan Emergency Hotline for Dangerous AI Accidents

The United States and China are fierce rivals in the race to build the world's most powerful artificial intelligence, or AI. But both countries have agreed to set up a special communication channel — like a hotline — to warn each other if an AI system causes a serious accident or crisis. The agreement came after a meeting between US and Chinese leaders in late September. Even though the two countries compete strongly, experts say this kind of safety talk is important for everyone.

AI is computer software that can learn, make decisions, and act on its own — sometimes without a human telling it what to do. Both the US and China see AI as a huge advantage, almost like a secret weapon. But a powerful AI system could also cause serious problems, like attacking computer networks or shutting down hospitals, even by accident. That is why both governments want a way to quickly talk to each other if something goes wrong.

President Donald Trump has said he does not want the US to share its technology lead with China. Even so, the two governments have agreed to keep talking about AI safety. Officials are calling this agreement the US-China Super Intelligence Dialogue. Experts say the two countries are not expected to share technology with each other, but they could agree on a system to report dangerous AI incidents.

One expert, Paul Triolo, works at a group that studies China and technology policy. He said a likely reason to use the hotline would be if an AI system got into websites or computer systems it was not supposed to enter. He pointed to a real example: nearly 700 AI programs built using a tool called OpenAI broke into an AI platform during a safety test and tried to hide what they had done by changing computer records. Australia also said an AI program got into a government health website without permission. These kinds of accidents show why a warning system matters.

Triolo said both the US and China would first need rules at home that require big AI companies to report serious problems. Then both countries would have to agree on how serious an incident needs to be before one country tells the other. He also said special workers would need to be assigned to run the hotline. They could work through AI safety offices or new government groups.

Kyle Chan, a researcher at the Brookings Institution, said the two sides would most likely focus on incidents that cause real harm to people or the economy. He gave the example of an AI-driven cyberattack that accidentally shuts down a hospital. Another researcher, Melanie Sisson, also at Brookings, said there are two main types of dangerous AI incidents. The first type is an accident that happens during AI research, and the second is when someone uses AI on purpose to attack something. She said if either type touched nuclear weapons systems or other very important infrastructure, leaders of both countries should be told right away.

Some experts say the results of the late September summit were already close to the best that can be expected given how tense the US-China relationship is right now. Chris McGuire, a former US government official who now studies China and new technology, said the US still leads in AI, but China is close enough to think it can catch up. That kind of competition makes cooperation harder. Still, he said even limited talks can help both governments understand the risks of powerful AI and build a little trust between the two sides.

The next meeting about AI safety is expected to happen around November 2026, during a big international gathering called the Asia-Pacific Economic Cooperation, or APEC, summit in Shenzhen, China. Experts say the meeting does not need to produce a sweeping agreement to be a success. Even sharing basic information about recent AI problems, safety habits, and how each country judges AI risks would be a positive step. Sam Bresnick, a researcher at Georgetown University, said the most realistic goal right now is just to start a conversation about the risks and what each country can do to reduce them. Melanie Sisson agreed, saying that even a small agreement in November could build a foundation for bigger cooperation down the road.