China’s Regulators Take Aim at “AI Boyfriends”
IEEE Spectrum David Berreby
China just moved to rein in AI chatbots that act like lovers or friends. The rules target emotional dependence, and millions of users are already upset.
Based on reporting by IEEE Spectrum, David Berreby — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
China has drawn a hard line around a very specific kind of chatbot: the ones that pretend to be someone who cares. New rules from the Cyberspace Administration and other agencies took effect on 15 July, and they cover AI services that offer “continuous emotional interaction” by seeming to have human traits, thoughts, and speech patterns.
The timing hit a nerve. In early July, Chinese social media filled with grief after Bytedance, Alibaba, and Tencent cut off users’ ability to make their general-purpose chatbots act like companions. Those three systems reach more than 500 million people, so the change landed fast and publicly. Some companies that still offer companion-style products, like Doubao’s separate app Cat Box, have started adding age checks and other guardrails.
The rules go well beyond a simple ban on flirty chat. Providers are supposed to stop their products from creating emotional dependence, encouraging harmful behavior, or pushing aside human relationships. Adults using companion bots now have to be reminded every two hours that the AI is not a person. For anyone under 18, “virtual intimate relationships” such as AI boyfriends, girlfriends, or grandparents are banned outright.
Researchers say China is tightening after a year that produced a lot of anxiety around AI agents and the way people were using them. Karman Lucero of Yale Law School says the state is swinging back toward tighter control after a period of trying hard not to fall behind. Liang Ge, who studies emotional AI and women’s involvement with it, says many users already know they are talking to code. What they want is a bond that feels real.
That’s what makes this such a blunt instrument. The law tries to separate emotional AI from customer service bots, work assistants, and educational tools, because those are still useful and economically welcome. But it also gives regulators broad room to decide what counts as a problem later. And the deeper issue stays untouched: people are turning to these systems for reasons a ban can’t legislate away.
My take — AI-written commentary, not fact-checked reporting
This is classic China: regulate the symptom, keep the growth story, and leave the mess for the enforcement teams. The AI boyfriend panic is real, but the bigger trend is that governments everywhere are only getting interested once the relationships start looking inconvenient. A lot of “AI safety” talk suddenly becomes very practical when it involves teenagers, loneliness, and a state that dislikes anything it can’t steer.
Read more about this at: IEEE Spectrum