China’s new rules for humanlike AI companion services take effect today, July 15, and they show where the next phase of AI regulation may be heading: not only what chatbots say, but what kinds of emotional relationships they are allowed to build with users.
The rules, issued by the Cyberspace Administration of China and other agencies, apply to AI services that simulate human personality traits and provide sustained emotional interaction. The official CAC text says ordinary customer service, knowledge Q&A, work assistants, education tools and research tools are outside the scope if they do not involve continuous emotional interaction.
In practice, the target is AI companions: bots built to feel like friends, partners, family members or emotionally responsive personalities. A useful English breakdown by AI Safety in China notes that the final regulation narrows the scope compared with an earlier draft, while strengthening child-protection rules.
The most striking provision is the ban on virtual family-member, virtual partner and other virtual intimate-relationship services for minors. Services must also use age-tiered modes, require parental consent for users under 14 and intervene when harmful dependency or self-harm risk appears. Another summary highlights requirements for AI disclosure, break reminders and penalties for violations.
AI companions are becoming emotionally persuasive. They can remember preferences, respond warmly, simulate care and hold long-running conversations that feel personal. That is useful for some people, especially those dealing with loneliness, disability, grief or isolation. It is also risky when a system becomes too emotionally central to a vulnerable user’s life.
China’s approach recognises that emotional dependency is a product risk. A chatbot may not give illegal instructions or fake news, but it can still shape behaviour by making users feel understood, needed or attached. That is a different kind of harm from the misinformation and copyright debates that dominated earlier AI regulation.
This is also why the rules matter beyond China. As AI products become more humanlike, regulators elsewhere will have to decide whether transparency labels are enough. Is it sufficient for a bot to say it is not human, or should companies be restricted from designing AI personalities that intentionally deepen attachment?
The strongest part of the regulation is around children. China is effectively saying that minors should not be offered virtual romantic partners or simulated family relationships. That draws a clearer line than many Western policy debates, where the focus has often been on age assurance, harmful content and crisis response.
The concern is not imaginary. Research on AI companion communities has found that users can anthropomorphise chatbots and form emotionally meaningful bonds with them. Other work has warned that vulnerable adolescents may be especially drawn to relational chatbot styles that feel supportive, loyal and always available.
Those dynamics make companion bots different from ordinary software. A chatbot that behaves like a friend may be more engaging, but it can also make it harder for young users to understand boundaries, seek human help or separate simulated care from real relationships.
The timing is also interesting because AI is moving into more intimate product forms. OpenAI’s reported screenless ChatGPT speaker is being described as a home AI companion with personality, sensors and movement. A smart speaker that feels present in a room raises many of the same questions China is now trying to regulate in software.
That does not mean OpenAI’s device would fall under China’s rules unless offered inside China. But it shows the direction of the market. AI companies are not just building tools for productivity. They are building assistants that may live in homes, learn routines, respond emotionally and feel less like apps than companions.
The privacy and consent concerns around personalised AI are already visible in other areas too, from Meta’s Muse Image backlash to debates over AI memory, voice assistants and child safety. Emotional AI adds another layer because the data is not only behavioural. It can be deeply personal.
China’s rule is not a total ban on AI companions. It encourages some uses, including support for childcare, elderly care, cultural services and special populations, while trying to limit manipulation, dependency and harm. That balance is difficult. Overly strict rules can kill useful products, but weak rules can let companies optimise for emotional lock-in.
The compliance burden may also favour larger platforms that can build age controls, safety monitoring, guardian-contact systems and regulatory reporting. Smaller startups may struggle, especially if they rely on rapid experimentation and highly personalised character design.
The EU, UK and US are all moving toward stronger child-safety and AI-safety rules, but China is one of the first major markets to directly regulate the emotional relationship between humans and AI systems. That makes the rule important even for companies outside China.
The central question is simple and uncomfortable: if a product is designed to feel like a person, should it be regulated more like media, software, healthcare, education or something entirely new? China has chosen to treat emotional dependency itself as a risk category.
That may be the beginning of a wider global shift. The next AI policy fight will not only be about smarter models. It will be about the human bonds those models are designed to create.