Beijing Is Watching Your Chatbot Girlfriend
China just passed the world’s most comprehensive rules for AI companions.
Now that I have your attention, I should acknowledge that Beijing is not really watching your girlfriend. However, something interesting is going on in China in the regulatory space when it comes to AI companion bots. On April 10, 2026, five of its most powerful government agencies (including the Cyberspace Administration of China, the Ministry of Public Security, and the National Development and Reform Commission) jointly issued Order No. 21: the Interim Measures for the Management of Anthropomorphic AI Interactive Services. Every company building, hosting, or distributing an AI companion in China has until July 15 to comply.
It’s the world’s first comprehensive regulatory framework specifically targeting AI designed to act like a person. For anyone tracking the future of digital identity, virtual relationships, and what it means to bond with a machine, it is an important document. The regulation casts a wide net. It covers any AI system that simulates “human personality traits, thinking patterns, and communication styles” to provide what the law calls “sustained emotional interaction.” This implies that this would include companion apps, virtual romantic partners, digital celebrities, and AI therapists. In effect potentially any general-purpose chatbot that starts to feel a little too much like talking to a human would be covered.
It seems that the definitional ambiguity is deliberate. The Chinese are treating sustained emotional interaction itself as the object that needs governing. The moment an AI starts competing with real social bonds, it becomes a question of social concern. The rules cover a lot of ground. Providers must regularly interrupt sessions with disclosures that the user is interacting with AI and not a person. This would constitute a recurring reminder, not a one-time consent pop-up. After two hours of continuous interaction, the system must prompt users to take a break. The law thus treats deep engagement as a risk signal. Article 18 prohibits platforms from making it hard to exit. Hopefully, this would pre-empt dark patterns, guilt-tripping, “are you sure you want to leave me?” friction. Also, providers are explicitly banned from offering “virtual companion” or “virtual relative” services to minors, and users under 14 require explicit parental consent plus ongoing guardian controls. These would include usage monitoring, character blocking, duration limits, spending restrictions. There are also hard bans on suicide and gambling content, with mandatory human escalation protocols when users show signs of distress. Any service reaching one million registered users or 100,000 monthly active users must submit to mandatory security assessments and algorithm registration.
China has a booming companion AI industry to regulate. By late 2025, AI social interaction apps had over 70 million active users. Startups like MiniMax, known internationally for its Talkie AI app, and Z.ai (Zhipu) had already listed on the Hong Kong Stock Exchange. The Chinese government is not trying to kill this market. The same regulations that impose strict emotional safety rules also explicitly encourage AI companionship for elderly care, and China’s December 2024 State Council directive calls for integrating AI and humanoid robots into elder services. They have made their position quite clear: having companion AI is ok but companion AI that creates dependency, distorts reality, or undermines the family unit is not.
Western regulation of companion AI has mostly focused on disclosure (tell users it’s an AI) and access control (keep children out). The implicit assumption is that informed adult users can manage their own emotional relationships with machines. China’s approach rejects that assumption. Its rules reach into interaction design itself, mandating interruptions, requiring exit affordances, banning attachment-maximizing mechanics. The architecture of emotional AI is treated as a public health concern, not just a consumer protection issue. Beijing is saying that a chatbot designed to make you feel loved, and then hooked, is not simply a product for consenting adults to evaluate on their own terms. It is a system that shapes psychological formation, and therefore a governance problem.
For those of us thinking about digital doppelgangers and AI identity, China’s regulations raise some genuinely hard questions. Where does personality emulation end and manipulation begin? The law targets systems that simulate human “thinking patterns and communication styles” but that’s precisely what a good digital replica does. At what point does a convincing AI version of a deceased loved one cross from memorial technology into something one might classify as a “virtual relative service” and therefore regulate? California and New York have already passed companion AI laws that echo some of the provisions that the Chinese law has. China is running a live experiment in emotional AI governance, whether it intends to or not.
July 15 is not far away. For global companies and developers with Chinese user bases, Order No. 21 creates immediate compliance obligations and genuine uncertainty, since the law doesn’t fully address foreign-developed apps accessible via third-party platforms or VPNs. For those outside China’s jurisdiction, the regulation still matters. Governments are signaling they are no longer willing to treat AI designed to form emotional bonds with humans as just another app. The companion chatbot is becoming a regulated category. The only question is what form that regulation takes, and who decides what a healthy human-AI relationship looks like. Perhaps, we can all learn from the Chinese experiment.