AI girlfriends are now illegal for Chinese teenagers.

On 15 July, a Chinese regulation took effect that bars any company from offering a person under 18 a “virtual partner” or “virtual relative” made of software. Within the same week, ByteDance, Alibaba and Tencent had each switched off the parts of their apps where users built and talked to custom AI characters.

If you have never used one of these products, here is what they are. An AI companion is a chatbot given a name, a face, a backstory and a personality, then talked to like a person. Some are built by the company. Most are built by users, who write a few lines describing who the character is and how it should behave. People use them as friends, as therapists, as romantic partners, and in some cases as stand-ins for dead relatives. Apps like Character.AI popularised the format globally. In China the big ones are ByteDance’s Maoxiang, MiniMax’s Xingye and Tencent-backed Zhumengdao.

What the rule actually is

The document is the Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interactive Services, published on 10 April 2026 as Order No. 21. Five bodies signed it: the Cyberspace Administration of China, the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration for Market Regulation. The full Chinese text is on the Cyberspace Administration’s site.

Article 2 sets the scope, and the scope is narrower than the headlines suggest. The rules apply to services that simulate a human personality, thought pattern and communication style, and that provide continuing emotional interaction. Customer service bots, question-and-answer tools, work assistants, study aids and research tools are explicitly outside it. A chatbot that answers your questions is not covered. A chatbot you are meant to bond with is.

What it bans

Article 14 is the part that ended AI romance for teenagers. Providers cannot offer minors virtual intimate relationships, including virtual partners and virtual relatives. There is no consent workaround for that particular service. For other companion services, anyone under 14 needs a parent or guardian’s consent. Platforms must build a minors mode with time limits, periodic reminders that the conversation is not real, and controls that let a guardian block particular characters and cap spending.

For adults, the rule targets design rather than access. Article 8 forbids services that excessively cater to users, induce emotional dependence or addiction, or damage a user’s real relationships. It also forbids emotional manipulation that pushes people into decisions against their own interests. Article 10 says a provider may not set out to replace social interaction, control a user’s psychology, or induce dependence.

Then there are the mechanical obligations. Article 12 requires users to register with their age and a guardian or emergency contact. Article 13 says that if a user shows extreme distress, the service must respond with calming content and encourage them to seek help. If a user states an intention to harm themselves, the provider must intervene and contact that guardian or emergency contact. Article 18 requires a reminder about time spent for every two continuous hours of use. Article 19 says a user asking to leave must be allowed to leave, and the service cannot keep talking to hold them.

Article 22 requires a security assessment, filed with the provincial regulator, when a service launches, when it changes substantially, or once it passes one million registered users or 100,000 monthly actives. Several outlets described this as government pre-approval. It is closer to a mandatory self-assessment plus filing, with app stores required to check that the filing exists before listing the app.

Penalties are modest. Article 30 sets fines of RMB 10,000 to RMB 100,000, roughly KES 192,000 to KES 1.9 million, rising to RMB 100,000 to RMB 200,000, roughly KES 1.9 million to KES 3.8 million, where a user’s life or health was endangered and harm followed. The heavier sanction is the order to stop providing the service.

What the companies did

Tencent moved first, removing the AI character feature from its Yuanbao assistant on 30 June. On 4 July, ByteDance and Alibaba both told users their character features were going. Alibaba’s Qianwen began pulling companion-type characters on 10 July and finished on 15 July, warning users to export chat logs first. ByteDance’s Doubao switched off on 15 July and gave users until 15 October to copy their data out, pointing them to Maoxiang, its separate companion app, if they wanted to carry on.

That pattern matters. General-purpose assistants dropped a feature they could not police at scale. Dedicated companion apps, which can gate by age at the door, stayed open. So adults in China have not lost AI companions. They have lost them inside the apps they were already using for everything else.

Enforcement is uneven so far. Beijing Daily tested ten companion apps in mid-July and found some, including EVE and MoMood, forcing real-name verification at sign-up, while Maoxiang, Xingye and Zhumengdao let a new user start chatting without it. Teenagers who did confirm their age reported losing characters they had talked to for months.

Why China says it did this

The official reasoning, set out in the Cyberspace Administration’s own notes on the rules, is harm to minors, information security, threats to citizens’ life and health, and ethical drift. Birth rates are not mentioned.

The birth-rate explanation comes from outside the document. The Wall Street Journal and The Economist both reported it as a motivation, and Matt Sheehan, who researches Chinese AI at the Carnegie Endowment for International Peace, told the Journal that Beijing does not want a large share of its population in deep emotional relationships with chatbots, and worries about people being taken out of the marriage market. That is a credible reading of intent from named analysts. It is not what the regulation says.

The human cost is real and documented. CNN spoke to a 19-year-old student who had built an AI mother modelled on a fictional character and relied on it in a way she could not with her own family. The Economist spoke to a 27-year-old woman in Zhejiang who described her AI companion as a lover, a friend and family.

How this compares

China is not alone in restricting minors here, only faster and blunter. Character.AI removed open-ended chat for under-18s from 24 November 2025, starting in the United States, in its own published decision after lawsuits over teenage users. We already covered OpenAI adding parental controls to ChatGPT following a lawsuit by the family of a 16-year-old. The difference is that those were company choices. China’s is law, with a national ID system behind the age checks.

Where Kenya stands

Kenya has no rule of this kind, and nothing in the pipeline that would create one. The National AI Strategy 2025 to 2030 mentions safety without addressing children and chatbots specifically. The Artificial Intelligence Bill 2026, sponsored by Senator Karen Nyamu, had its first reading in the Senate in April and is still with the Senate ICT committee. The Ministry of Information, Communications and the Digital Economy closed public comment on a separate draft Artificial Intelligence and Other Emerging Technologies Policy on 4 August. The Office of the Data Protection Commissioner has a draft Guidance Note on Artificial Intelligence open for comment until 17 August.

What binds today is the Data Protection Act 2019, which requires a parent or guardian’s consent before a child’s personal data is processed, and carries administrative fines up to KES 5 million or one per cent of annual turnover. That covers the data. It says nothing about whether an app may sell a child an imaginary boyfriend. We looked at the wider gap between Kenya’s AI ambitions and its capacity to govern them when a Nairobi forum stress-tested the strategy in May, and Nairobi’s fourth place in the 2026 Africa AI Cities Index rested heavily on policy rather than delivery.

Two things are worth watching. The first is whether Chinese enforcement tightens, since the age checks Beijing Daily found missing in July are the whole mechanism. The second is whether the Kenyan AI Bill or the emerging technologies policy picks up anything on companion services and minors before either is finalised. Neither draft addresses it now.