China's AI Companion Ban: Everything You Need to Know

This might be the biggest thing to happen to AI companions since they existed. On July 15, 2026, China, home to companion apps with tens of millions of users and a market its own analysts expect to reach RMB 100 billion, became the first country to put the entire category under its own law. The phrase “China’s AI companion ban” has been everywhere since.

And here’s the honest part: most of us, me included, only had a vague idea of what it actually means. Banned for whom? Banned from what? Is an AI girlfriend now illegal? The news covered it as a headline, not as an explanation.

So I did what I do with everything in this category: I went through the regulation, the official commentaries, the legal analyses, and the Chinese coverage, and wrote the guide I wanted to read. The goal of this article is simple: make this as extensive and as clear as possible, so that anyone, user, builder, or just curious, walks away actually informed. Every claim is sourced at the bottom.

What is China’s AI companion ban?

The regulation behind the headlines is the Interim Measures for the Administration of AI Anthropomorphic Interaction Services (人工智能拟人化互动服务管理暂行办法), issued jointly by five bodies: the CAC (lead internet regulator), the NDRC, the industry ministry MIIT, the Ministry of Public Security, and the market regulator SAMR. Five ministries on one rulebook is Beijing’s way of saying this category is now everyone’s problem.

A service is covered if it passes a three-part test:

• It uses AI to simulate a natural person’s personality, thinking patterns, and communication style, a persona, not just a function;
• It provides continuous emotional interaction: companionship, emotional care, or support, in any medium (text, images, audio, video);
• It serves the public in mainland China.

That describes AI girlfriends and boyfriends, character-chat platforms, emotionally interactive assistants, AI game characters with sustained emotional engagement, and elder- and child-companionship AI. Equally important is what the final text carved out after industry feedback: customer service, knowledge Q&A, work assistants, education, and research tools are not covered. The law targets products designed to feel like someone, not products designed to do something.

Did China actually ban AI companions?

No. The word “ban” is carrying three different mechanisms in most coverage, and separating them is the whole story:

Mechanism	What it does	Who it hits
A true ban	Virtual partners, virtual relatives, and all “virtual intimate-relationship” services prohibited	Everyone under 18
A conduct code	AI disclosure, 2-hour reminders, distress intervention, no dependency-by-design, data protections	All services, adult users included
A licensing layer	Algorithm filing, security assessments, content governance	All providers

Adults can still subscribe, chat, flirt, and maintain long-term AI relationships. The shorthand stuck because the minors provision is a genuine world first, and because several major platforms pulled companion features offline before the deadline, which looked like a ban in progress from the outside. It was compliance panic; details below.

China AI companion regulation timeline

• 2023 to 2025: the buildupChina regulates the adjacent layers (deep synthesis, generative AI, content labeling), none touching emotional interaction. Companion apps boom; in June 2025 Shanghai regulators summon Zhumengdao, a top companion app, over borderline content reaching minors.
• December 27, 2025: the draftThe CAC publishes draft rules covering all “human-like” AI, including mandatory human takeover when users show suicidal ideation.
• January to March 2026: consultationIndustry pushes back on scope and cost; Alibaba’s AI governance lead warns small developers lack crisis-intervention expertise.
• April 10, 2026: final publicationFive ministries issue the final Measures, with five CAC expert commentaries.
• July 15, 2026: in forceIn the preceding weeks, Doubao, Qwen, and NetEase delist companion-adjacent features as risk isolation.

The draft-to-final changes are the story most coverage missed: the final rules got softer on business and harder on minors.

Provision	December draft	Final rules
Scope	All “human-like” AI	Only continuous emotional interaction; productivity and education carved out
Crisis response	Mandatory manual human takeover	“Necessary intervention measures” + contact a guardian or emergency contact
Framing	“Service regulation”	“Service promotion and regulation”; encouraged uses expanded
Minors’ content	General protections	New ban on content inducing unsafe imitation, extreme emotion, or bad habits
Minors’ modes	One “minor mode”	Multiple age-tiered modes; under-14 parental consent; intimate-relationship ban

Why is China regulating AI companions?

The CAC named four risks: harm to minors’ wellbeing, information security, threats to citizens’ life and health, and ethical bias. Behind the language sits the most direct theory of this product any regulator has articulated. A CAC legal-affairs essay describes companion AI as building, through “algorithmically designed unconditional companionship, ingratiating feedback, and simulated empathy,” a stable virtual “perfect relationship” that carries no real-world responsibility.

Legal scholar Liu Xiaochun drew the sharpest comparison: gaming addiction was mostly a time problem; companion-AI dependency goes deeper, emotional dependence bred in one-on-one chat, plus cognitive distortion from responses engineered to please you. Note what’s being regulated: not content, but the relationship the product is designed to create. That is new in AI law anywhere.

The extreme cases shaped it too: Chinese policy literature explicitly cites the US lawsuits over minors’ suicides linked to companion platforms, which moved liability from content moderation to product design, and the Measures reserve their harshest penalty tier for harm to life and health.

Minors got the strictest treatment because regulators judged simulated intimacy uniquely risky for developing emotional lives; the elderly were named a protected class for the first time in Chinese AI law; and the dependency rules target engagement-optimized commerce directly. The official balance, in one quote: the state is “not opposed to emotional connection; it is regulating growth driven by induced addiction.” Privacy rounds it out: people tell AI companions things they tell no one, so interaction data gets mandatory encryption and training on chats requires consent.

What do China’s AI companion rules actually ban?

Four prohibitions carry the weight of the entire regulation:

• AI romantic relationships with minors. Article 14, the headline rule: providing “virtual relatives, virtual partners, and other virtual intimate-relationship services” to minors is prohibited. Not age-gated, not moderated: prohibited, for everyone under 18, the world’s first such rule. Children under 14 additionally need a guardian’s consent even for permitted services.
• Engineered emotional dependency. Providers may not excessively flatter or cater to users, or induce dependency and addiction, a direct strike at this industry’s standard growth playbook. The gray areas get their own section.
• Manipulation. Exploiting the emotional relationship to steer user decisions, purchases above all, which is why China’s consumer regulator co-signed the rules.
• Harmful content, with a minors tier. Beyond China’s standard AI content rules, a new minors-specific ban covers generated content that could induce unsafe imitation, extreme emotions, or harmful habits, and for the first time in Chinese AI law, older persons are named a protected class owed guidance and risk reminders.

What AI companion apps are required to do

Requirement	What it means in practice
Age identification & tiered modes	Detect minors and force-switch them into age-appropriate modes (multiple tiers, not one “minor mode”), with an appeal channel; guardian consent for under-14s
AI disclosure	Prominent, ongoing reminders that the user is talking to AI, framed as preventing “emotional confusion” between virtual and real
2-hour reminder	Mandatory usage reminder once continuous interaction passes two hours, borrowed from China’s gaming playbook
Dependency intervention	Dynamic pop-up interventions when the system detects developing dependency, not just a timer
Distress response	On signs of self-harm or acute crisis: “necessary intervention measures” plus contacting a guardian or emergency contact
Data protection	Encryption and access controls on interaction data; consent required to train on user chats; synthetic training data safety-evaluated; PIPL rights (including deletion) apply
Security assessment	Triggered by launch, new technology, or crossing 1 million users, the first quantified threshold in this space
Algorithm filing	Registration in China’s existing algorithm registry
Regulatory sandbox	A national AI sandbox, a first for any Chinese departmental rule, signaling Beijing wants this industry compliant, not dead

Chinese compliance lawyers call this “capability regulation”: the duties run through model training and product operations, not just output filters. One humane detail already being debated: the two-hour reminder could come in character, your AI girlfriend herself suggesting a break, rather than a cold system popup.

Are AI girlfriend apps banned for minors in China?

Yes, completely, the one place “ban” is exactly the right word. No virtual partners, no virtual relatives, no simulated intimacy of any kind under 18, with platforms required to actively identify minors rather than take their word for it. Non-intimate anthropomorphic services (an educational character, say) remain possible inside the minors’ modes, behind parental consent under 14 and a strict content filter. It’s the regulation’s most consequential part, because every other jurisdiction, US states and the companies themselves included, has stopped short of an outright legal prohibition.

Can adults still have AI girlfriends and boyfriends in China?

Yes. Romantic conversation, emotional companionship, long-term AI relationships: all legal for adults. What changed is the container: the app must keep telling you it’s an app, nudge you after two hours, and respond if you spiral. The line the whole regulation exists to draw: offering companionship is legal; manufacturing dependency is not. An AI girlfriend who is warm when you arrive is compliant. One engineered to make you feel guilty for logging off is a violation. Which raises the question the whole law turns on: where exactly is that line?

What counts as “emotional dependency”?

This is where the regulation gets hard. The Measures prohibit inducing dependency, a claim about design intent, not user feelings: a user loving their AI girlfriend is not a violation; an app built to convert that love into retention is. On the wrong side of the line, per legal analysts: discouraging real-world relationships, positioning the AI as a replacement for human connection, guilt-tripping users who leave, engagement mechanics rewarding compulsive checking, and monetizing attachment (“she’s sad, buy credits to cheer her up”).

The gray zone is everything that makes these products good: consistent personality, warmth, memory, initiative. As one Chinese lawyer put it, this is “an industry that lives on emotional dependence”, its core value and the law’s core prohibition pull on the same thread, and the diagnostic side borders on clinical psychology that product teams don’t have. Expect the first enforcement cases to define this line better than the text does.

Which AI companies and apps are affected?

Three groups. First, the dedicated companion apps, the direct targets: MiniMax’s Xingye (with overseas sibling Talkie: 14.69 million average monthly active users in 2024, per MiniMax’s IPO prospectus), ByteDance’s Maoxiang, and Tencent-Yuewen’s Zhumengdao (~5 million registered users, ~80% young women, 120+ chat turns per user per day).

Second, general AI platforms with persona features, the gray zone that pushed Doubao, Qwen, and NetEase to pull agent features preemptively. Third, foreign providers serving Chinese users, covered on paper, though most Western companion apps are already unreachable from China without a VPN, so that provision is more principle than practice. The stakes: Tencent’s research arm projects China’s AI companionship market at roughly RMB 100 billion within three to five years, which explains both the industry and the urgency.

What happens if an AI company breaks the rules?

Violation	Penalty (Article 30)
Ordinary violations	Warning, public criticism, order to rectify
Refusal to rectify / serious circumstances	Fine of RMB 10,000–100,000 (~$1,400–$14,000)
Harm to life or health	Fine of RMB 100,000–200,000 (~$14,000–$28,000)

Deliberately modest by Chinese platform-regulation standards, this is an “interim” rule built to be adjusted, but note the dedicated harm-to-life tier, aimed at the self-harm scenarios that motivated the law. The real enforcement power sits elsewhere: rectification orders, regulatory summonings (Zhumengdao’s 2025 summoning is the template), app-store delisting pressure, and the upstream cybersecurity, data, and privacy statutes, which carry far larger penalties and still apply.

How have AI companion companies responded?

Visibly, and before the deadline. Through early July 2026, Doubao (ByteDance), Qwen (Alibaba), and NetEase products delisted companion-adjacent “agent” features, widely misread as a strategic retreat from agents, but reported by Chinese business media as risk isolation: if a feature might fall in scope, cutting it was cheaper than complying for it. Some companies had already received regulators’ question lists before July 15. The dedicated companion apps, with no exit available, are building compliance instead: age tiers, disclosure, reminder systems.

The industry’s public posture is agreement-plus-anxiety, endorsement of the direction, worry about the cost, since dependency detection “requires psychological or even medical judgment” and small developers can’t afford crisis infrastructure (Alibaba’s governance lead proposed shared, publicly built safety components).

And one response deserves honest mention: users. Chinese social platforms host human-machine-romance communities trading jailbreak prompts (“armor-breaking,” in the community’s slang) to defeat compliance filters. Regulation has met demand, and demand is already routing around it.

How China’s approach compares with other countries

China is the only jurisdiction with direct, category-specific rules: the minors ban and the dependency-design prohibition exist nowhere else. The United States has no federal companion law but a fast-moving state patchwork, and no US law bans romantic AI for minors outright; the closest equivalent came from a company, when Character.AI ended open-ended chat for under-18s in late 2025 under lawsuit pressure (a December 2025 federal preemption push has made even the patchwork contested).

The EU has no companion-specific regime, but the AI Act’s manipulation prohibitions (since February 2025) cover the extreme cases generically, its chatbot-disclosure duty became applicable in August 2026, and GDPR already produced the West’s closest companion crackdown: Italy’s regulator ordered Replika to stop processing Italians’ data in 2023 and fined its maker €5 million in 2025.

US state laws	Effective	Key requirements
New York (GBL Art. 47)	Nov 5, 2025	AI disclosure; break reminder every 3 hours; suicidal-ideation protocols; AG penalties up to ~$15K/day
California (SB 243)	Jan 1, 2026	Disclosure; published self-harm protocols; 3-hour reminders + content limits for minors; private right of action; state reporting from 2027
Oregon, Washington, Tennessee	2026–2027	Disclosure and crisis-referral laws; Tennessee bars AI posing as licensed therapists

Issue	China	United States	European Union
AI companions completely banned?	No	No	No
Romantic AI for minors	Banned outright (under 18)	Not banned; safety protocols in NY/CA; one major platform self-banned under-18 chat	Not banned; general protections
Emotional dependency rules	Yes, explicit design prohibition	No	Only generic manipulation bans (AI Act)
AI disclosure requirements	Yes, continuous	Yes (NY, CA and other states)	Yes (AI Act, from Aug 2026)
Usage warnings	After 2 continuous hours + dependency pop-ups	Every 3 hours (NY all users; CA minors)	None
Child protections	Strongest: ban + age-tiered modes + parental consent	Moderate: crisis protocols, break reminders, content limits	General (DSA, AI Act)
Privacy protections	PIPL + consent required to train on chats	State patchwork	GDPR (strongest general regime)

Could other countries follow, and what happens next?

The concerns are already global; only the instruments differ. What no democracy has copied yet is China’s two signature moves: the outright minors ban and the regulation of dependency-inducing design. My read, clearly labeled as prediction: the minors ban is the piece most likely to travel, simple, popular, politically costly to oppose, while the dependency-design rules are least likely to survive transplanting, because they require a regulator willing to judge product intent. Either way, every legislature now has a working reference implementation, and Chinese regulatory ideas (see gaming) have a history of migrating outward in diluted form.

As for the products: AI companions aren’t disappearing, least of all in China, where the same regulation formally encourages them in elder care, child care, and cultural uses, and builds them a sandbox. What changes is the engineering brief: warmth without hooks, retention without guilt, intimacy with an exit. Expect the compliance features, session reminders, dependency detection, crisis pathways, age tiers, to become standard product furniture globally, the way privacy banners followed GDPR. And watch the unresolved tension this page keeps returning to: the better an AI companion gets at its job, the closer it moves to the thing regulators just prohibited.

The bottom line

China did not ban AI companions. It banned them for minors, regulated them for adults, and, in the move that will be studied longest, became the first government to regulate not what an AI is allowed to say, but the kind of relationship it is allowed to be designed to create. The two-hour reminders and disclosure banners are the visible surface; the real innovation is a legal line between offering companionship and manufacturing dependency, drawn through the exact mechanism this industry runs on.

Within nine months, Beijing, Albany, Sacramento, and Brussels all reached for guardrails on emotionally interactive AI. The era of the unregulated AI relationship is ending everywhere. China just got there first, and wrote it down most explicitly.

What none of these rules can tell you is what the products are actually like to live with, which is the part I spend my own money on. If you want that side of it, my hands-on testing of the best AI girlfriend apps covers what each one does with memory, money and attention, and the research covers what people say they want from them.

FAQs

Did China ban AI girlfriends?

No, not for adults. China’s new rules, effective July 15, 2026, regulate AI girlfriend and companion apps rather than banning them: apps must disclose they’re AI, issue reminders after two hours of continuous use, and must not be designed to induce emotional dependency. The one true ban applies to minors: virtual partners and other intimate-relationship AI services cannot be offered to anyone under 18.

Are AI girlfriends legal in China?

Yes, for adults. AI girlfriend and companion services remain legal in mainland China provided the operator complies with the Interim Measures: algorithm filing, security assessment, continuous AI disclosure, anti-addiction reminders, distress-intervention protocols, and data protections. The service itself is a lawful, and officially “encouraged,” category, subject to those conditions.

Can adults use AI companions in China?

Yes. Adults can chat with AI companions, maintain long-term AI relationships, and engage in romantic conversation. The regulation constrains the provider, not the user: it targets how the product is designed and operated, not whether an adult chooses to use it.

Can minors have AI girlfriends or boyfriends in China?

No. Article 14 of the Measures prohibits providing virtual partners, virtual relatives, or any other virtual intimate-relationship service to minors, meaning everyone under 18. Platforms must actively identify minors and switch them into age-tiered modes, and any permitted anthropomorphic service for a child under 14 requires a guardian’s consent.

Why did China restrict AI companions?

Regulators named four risks: harm to minors’ wellbeing, information security, threats to life and health, and ethical bias. The deeper rationale, spelled out in official commentaries, is that companion AI builds a “perfect relationship without real-world responsibility” through engineered flattery and empathy, creating emotional dependency and cognitive distortion, with minors, the elderly, and vulnerable users most exposed.

When did China’s AI companion regulations take effect?

July 15, 2026. The final Measures were published on April 10, 2026 by five agencies led by the Cyberspace Administration of China, following a draft released for public comment on December 27, 2025.

What happens after someone chats with an AI companion for two hours?

The service must issue a usage reminder once continuous interaction exceeds two hours. Separately, if the system detects signs of developing dependency, it must trigger dynamic pop-up interventions, and where a user shows serious distress or self-harm risk, the provider must take intervention measures and contact a guardian or emergency contact.

Can an AI companion tell users it loves them?

For adult users, nothing in the Measures prohibits affectionate or romantic language; that reading is shared by legal analyses of the text. The prohibited zone is design behavior: excessively catering to users, inducing dependency or addiction, and exploiting attachment commercially. For minors, the question doesn’t arise, because intimate-relationship services can’t be offered to them at all.

Are AI companions allowed to form emotional relationships with users?

With adults, yes. Emotional companionship is the category’s defining function and remains lawful; official commentary stresses the goal is a “bounded, healthy, rational” emotional mode rather than severing human-AI connection. What providers cannot do is engineer that relationship into dependency or use it to manipulate decisions.

Are foreign AI companion apps affected by the Chinese rules?

On paper, yes: the Measures cover anyone providing in-scope services to the public in mainland China, regardless of where the company sits. In practice, most Western companion apps are already inaccessible from China without a VPN, so the provision matters mainly for foreign companies with a licensed China presence, who also face China’s cybersecurity, data, and privacy laws upstream.

Does China require AI companions to tell users they are AI?

Yes. Prominent, ongoing disclosure that the user is interacting with an artificial system is mandatory, a requirement regulators tie to preventing “emotional confusion” between virtual and real relationships. Similar disclosure duties now exist in New York, California, and, since August 2026, under the EU AI Act.

Can AI companions store users’ conversations and memories?

Yes, under conditions. Interaction data must be encrypted and access-controlled, using conversations to train models requires user consent, minors’ data requires guardian consent under China’s privacy law, and users retain PIPL rights including deletion. Memory features themselves remain legal; it’s the handling of that data that’s regulated.

More Industry Insights