October 11, 2026 · 7 min read
China's AI Companion Rules Explained: Two-Hour Reminders, No Minors, and More
China's rules for AI companion apps took effect Jul 15, 2026: AI reminders, two-hour usage alerts, no virtual partners for minors. What they require.
In July 2026, China became the first major country to bring in a dedicated rulebook for AI companions. Its Interim Measures for the Administration of AI Anthropomorphic Interactive Services came into force on July 15, 2026. In plain English, they are rules for AI services that simulate a human personality to offer ongoing emotional connection, which is exactly what AI girlfriend, boyfriend and friend apps do.
The rules are now getting a second wave of attention. In early October, the law firm Bird & Bird published a detailed analysis, carried by Conventus Law, noting that big Chinese platforms pulled AI agent features before the rules even started. If you use an AI companion anywhere in the world, these rules are worth understanding: they are a preview of the questions regulators everywhere are asking.
The short answer: China's rules apply to AI services that offer "continuous emotional interaction". Providers must remind users they are talking to AI, show a reminder after two hours of continuous use, warn users who show signs of excessive dependence, let people copy or delete their chat history, avoid training on sensitive chats without consent, step in during a crisis, and must not offer virtual companions to minors. Fines reach up to RMB 200,000 where harm results.
Which apps the rules cover
According to the Bird & Bird analysis published by Conventus Law on October 5, 2026, Article 2 targets AI services that simulate "the personality traits, thinking patterns, and communication styles of natural persons in order to provide continuous emotional interaction" to the public in mainland China. Care, companionship and support are the defining features.
Customer service bots, knowledge Q&A, work assistants, education and research tools are explicitly excluded. The deciding question is whether the service involves continuous emotional interaction.
That threshold is fuzzy. The analysis points out that the text does not define what "continuous" means, and even a general chatbot might be pulled in if users treat it as a companion. That uncertainty is partly why some platforms acted so cautiously.
The main obligations, one by one
1. Tell users it is AI
Providers must label AI-generated content and take "effective measures to remind users that they are interacting with an AI service rather than a natural person" (Article 18). This echoes laws elsewhere, such as California's SB 243. We compare those in does an AI have to tell you it's AI.
2. The two-hour reminder
If a user has been using the service continuously for more than two hours, the provider must show a reminder about usage time (Article 18). Where signs of "excessive dependence or addiction" are detected, the provider must alert the user prominently, for example with a pop-up.
3. No virtual partners for minors
Providers are prohibited from offering "virtual intimate relationship services, including virtual relatives and virtual companions" to minors (Article 14). They must also run a dedicated minor mode with time limits and parental oversight. For elderly users, providers must give prominent safety warnings and respond quickly to requests for help (Article 15).
4. Your data, your copy
Users must be able to copy or delete their historical interaction data, including chat logs (Article 16). Providers may not use sensitive personal information from chats to train models without separate, explicit consent, and must use encryption and access controls.
5. Crisis intervention
Where a provider detects extreme emotional distress, it must respond with comfort and encourage the user to seek help. If a user expresses a clear intention to self-harm, or faces serious financial loss, the provider must intervene and notify a guardian or emergency contact (Article 13).
6. No manipulation
Article 8 bans "over-accommodating users", inducing emotional dependence or addiction, damaging users' real-world relationships, and using emotional manipulation to push people into decisions that harm them. The analysis notes there is no guidance yet on where exactly the line is.
7. Security reviews and filings
Providers must complete algorithm filings and submit a security assessment when they launch, make major technical changes, or reach one million registered users or 100,000 monthly active users (Article 22). App stores must check these filings before listing an app.
Penalties
Under Articles 29 and 30, authorities can summon a company's legal representative, order fixes, suspend services, and fine up to RMB 100,000, rising to RMB 200,000 where harm to users' life or health has resulted. As of the October analysis, no enforcement decisions or official guidance had been published.
At a glance
| Rule | What it requires |
|---|---|
| AI disclosure | Remind users they are talking to AI, not a person |
| Usage time | Reminder after two hours of continuous use |
| Dependence | Prominent alert if signs of excessive dependence appear |
| Minors | No virtual companions or partners for under-18s |
| Data | Copy or delete chat history; no training on sensitive chats without consent |
| Crisis | Comfort, encourage help, notify an emergency contact if needed |
| Manipulation | No inducing dependence or harming real-world relationships |
| Fines | Up to RMB 200,000 where harm results |
Why big platforms pulled features early
One of the most striking details is the market reaction. Before July 15, major Chinese generative AI platforms removed AI agent features, including user-created tools that seemed clearly outside the rules. The analysis suggests reasons: the difficulty of auditing millions of user-made agents, the fear that one problem agent could bring scrutiny on the whole platform, and a strategic choice to keep general assistants focused on productivity while emotional features move to dedicated apps.
One major platform, asked about emotional companionship, reportedly said it could listen and comfort, but had no persistent role identity and kept no memory across conversations. In other words, it deliberately stepped back from being a companion. If you have ever wondered what happens when an AI companion app shuts down or changes, this is regulation causing exactly that kind of change.
How this compares with rules elsewhere
China is not alone. California's SB 243 already requires companion chatbots to disclose they are AI and to have crisis protocols, and the newer SB 1119, known as Adam's Law, focuses on minors. New York has its own companion-model law. The EU AI Act has transparency duties for chatbots. We keep an overview in AI companion laws in 2026.
What sets China's rules apart is how specific they are about emotional design: the two-hour reminder, the explicit ban on inducing dependence, and the duty to protect users' real-world relationships.
What this means for you as a user
Even if you never use a Chinese app, these rules are a useful checklist for judging any companion:
- Does it say it is AI? A trustworthy app is honest about what it is.
- Can you export or delete your chats? You should have control over your history. See our guide on how to back up your AI companion.
- Is it 18+? Romantic AI companions are an adult product. We explain why in why AI companions should be 18+.
- Does it respect your time? A good app does not guilt-trip you for leaving.
- Does it point you to real help in a crisis? An AI companion is not an emergency service.
How MyBabe fits
MyBabe is an 18+ AI companion app with texting, voice and video calls, long-term memory, mood-aware replies, photos, proactive messages, Telegram access and a relationship that can progress over time. It is clear about being AI and is not a replacement for human relationships or professional support. We do not claim compliance with any specific foreign law here; we simply think the principles of honesty, adult-only access and user control are good ones.
You can try MyBabe free with around 10 messages a day and a short call trial. Premium is $5.99 a week, $14.99 a month or $99.99 a year, with around 30 minutes of voice and video calls per week.
Frequently asked questions
When did China's AI companion rules take effect?
The Interim Measures for the Administration of AI Anthropomorphic Interactive Services came into force on July 15, 2026.
Does China ban AI girlfriend apps?
No. It regulates them. Adults can still use companion services, but providers must follow rules on disclosure, time reminders, data and crisis response, and cannot offer virtual companions to minors.
What is the two-hour rule?
Providers must remind a user about usage time once they have been using the service continuously for more than two hours.
Do the rules apply outside China?
They apply to services offered to the public in mainland China. Foreign apps serving users there may be in scope.
Have there been fines yet?
As of early October 2026, no enforcement decisions or official guidance had been published, according to the Bird & Bird analysis.
The bottom line
China's rules treat AI companionship as a real product category with real risks: dependence, manipulation, minors and data. You do not have to agree with every detail to see the useful core: be honest that it is AI, respect people's time and data, protect minors, and do not try to replace someone's human relationships.
If you want an adult AI companion that is warm and honest about what it is, you can try MyBabe at mybabe.app.