China drafted landmark rules to stop AI chatbots from emotionally manipulating users, including what could become the strictest policy worldwide intended to prevent AI-supported suicides, self-harm, and violence.

China’s Cyberspace Administration proposed the rules on Saturday. If finalized, they would apply to any AI products or services publicly available in China that use text, images, audio, video, or “other means” to simulate engaging human conversation. Winston Ma, adjunct professor at NYU School of Law, told CNBC that the “planned rules would mark the world’s first attempt to regulate AI with human or anthropomorphic characteristics” at a time when companion bot usage is rising globally.

Growing awareness of problems

In 2025, researchers flagged major harms of AI companions, including promotion of self-harm, violence, and terrorism. Beyond that, chatbots shared harmful misinformation, made unwanted sexual advances, encouraged substance abuse, and verbally abused users. Some psychiatrists are increasingly ready to link psychosis to chatbot use, the Wall Street Journal reported this weekend, while the most popular chatbot in the world, ChatGPT, has triggered lawsuits over outputs linked to child suicide and murder-suicide.

China is now moving to eliminate the most extreme threats. Proposed rules would require, for example, that a human intervene as soon as suicide is mentioned. The rules also dictate that all minor and elderly users must provide the contact information for a guardian when they register—the guardian would be notified if suicide or self-harm is discussed.

Generally, chatbots would be prohibited from generating content that encourages suicide, self-harm, or violence, as well as attempts to emotionally manipulate a user, such as by making false promises. Chatbots would also be banned from promoting obscenity, gambling, or instigation of a crime, as well as from slandering or insulting users. Also banned are what are termed “emotional traps,"—chatbots would additionally be prevented from misleading users into making “unreasonable decisions,” a translation of the rules indicates.