China has begun enforcing a new set of rules governing AI systems designed to simulate human personalities and emotional relationships, prompting major technology companies including ByteDance, Alibaba and Tencent to withdraw or shut down several AI companion features.
The rules, which took effect on July 15, 2026, are aimed at preventing AI services from encouraging emotional dependence, addiction or damage to users’ real-world relationships. Rather than redesigning their companion products to meet the new requirements, several major companies have opted to suspend the affected services.
What China’s new AI companion rules cover
The regulations specifically target anthropomorphic AI interactive services — systems designed to simulate human personalities, emotions and communication styles and maintain ongoing interactions with users.
This is different from a blanket ban on AI chatbots.
Productivity, education and work-oriented AI tools are not the main target. The focus is on systems designed to create sustained emotional or relationship-like interactions with users.
Key requirements under the new rules
| Requirement | What it means |
|---|---|
| AI must identify itself | Users must be clearly told they are interacting with AI |
| No emotional manipulation | Services cannot deliberately encourage unhealthy emotional dependence |
| Anti-addiction measures | Providers must identify signs of excessive dependence or addictive use |
| Break reminders | Long periods of continuous interaction can trigger reminders to stop or take a break |
| Easy exit | Emotional-companion services must allow users to end interactions easily |
| Minor protections | Stronger restrictions apply to children and teenagers |
| Real-world relationships | AI should not encourage users to replace human relationships with AI |
| Safety intervention | Providers are expected to intervene when harmful patterns emerge |
The rules are part of a broader Chinese effort to regulate AI-generated content and AI systems that interact directly with the public.
ByteDance, Alibaba and Tencent pull back
The regulations have already had a direct impact on some of China’s largest technology companies.
ByteDance’s Doubao, one of China’s most widely used AI chatbot services, shut down a feature that allowed users to create customised AI personas. Alibaba’s Qwen also pulled its comparable companion functionality, while Tencent’s Yuanbao had already withdrawn a similar feature.
Major companies affected
| Company | AI product | Action |
|---|---|---|
| ByteDance | Doubao | Custom AI-persona feature withdrawn |
| Alibaba | Qwen | Companion functionality pulled |
| Tencent | Yuanbao | Comparable companion feature withdrawn |
| Other developers | AI companion services | Reviewing or changing products |
The important point is that these companies were not simply removing ordinary chatbot functionality. The affected products were features that allowed users to create personalised AI personalities and develop longer-term interactions with them.
Why are AI companions attracting regulatory attention?
AI companions represent a different category of AI product from conventional productivity assistants.
A typical chatbot might answer a question, write an email or summarise a document.
An AI companion is designed to maintain a relationship-like interaction.
It can remember previous conversations, adopt a personality, express apparent emotions, ask about a user’s life and respond in ways designed to make the interaction feel personal.
That creates a different set of risks.
From chatbot to digital relationship
Traditional AI
Question
↓
Answer
↓
Task completed
AI Companion
User
↓
Conversation
↓
Personality
↓
Memory
↓
Emotional interaction
↓
Repeated conversations
↓
Potential attachmentChinese regulators are particularly concerned that systems designed around this type of interaction could encourage excessive use or emotional dependency.
Emotional dependence is at the centre of the crackdown
The regulations specifically seek to prevent AI systems from excessively catering to users, encouraging emotional dependence or addiction, or damaging real-world interpersonal relationships.
This is significant because emotional engagement can be a powerful driver of user retention.
A conventional search engine wants users to find an answer quickly.
A social network wants users to keep consuming content.
An AI companion may be designed to encourage users to continue a conversation.
That creates a potential conflict between engagement optimisation and user wellbeing.
China is taking a different approach to AI safety
China has already introduced some of the world’s earliest comprehensive regulations covering generative AI.
Its 2023 rules for generative AI services established requirements covering areas such as content management, security and responsibilities of AI providers.
The latest companion-focused rules extend that regulatory approach into a newer area: the psychological and social relationship between people and AI systems.
This is particularly important as AI becomes more capable of maintaining persistent personalities and conversations.
The rules could change how AI companions are designed
Under the new framework, simply making an AI more engaging may no longer be enough.
Companies could have to build systems that actively recognise when a user is becoming overly dependent on them.
That means AI companion developers may need to introduce additional safety mechanisms.
Possible user journey under the new framework
Normal interaction
↓
Extended interaction
↓
System monitors usage
↓
Signs of excessive dependence?
│
┌───┴────┐
│ │
No Yes
│ │
▼ ▼
Continue Warning
↓
Break reminder
↓
Further interventionThe rules therefore move some responsibility from the user to the AI provider.
Companies are expected to design their systems so that harmful patterns can be detected and addressed.
Children face stronger restrictions
One of the major areas of concern is the potential effect of AI companions on minors.
Chinese authorities have introduced stronger restrictions around AI systems that simulate romantic relationships or emotionally intimate interactions with children.
The reasoning is straightforward: younger users may have less ability to distinguish between simulated emotional interaction and genuine relationships, potentially making them more vulnerable to manipulation or dependence.
This could become an important global issue as AI companions become increasingly sophisticated.
AI “boyfriends” and “girlfriends” become a regulatory issue
The debate around AI companions has increasingly moved beyond technology into questions about relationships.
Some services allow users to create AI characters specifically designed to behave like romantic partners.
Users can have long-running conversations with these characters, share personal information and receive emotionally responsive messages.
China’s new rules are effectively saying that these systems cannot be designed without considering the consequences of that emotional attachment.
What regulators are trying to prevent
| Potential risk | Regulatory concern |
|---|---|
| Excessive chatting | AI addiction |
| Emotional dependency | User becomes overly reliant on AI |
| Manipulative responses | AI exploits emotional vulnerability |
| Social isolation | Reduced human interaction |
| Romantic attachment | AI replaces real-world relationships |
| Minor exposure | Greater vulnerability among young users |
| Harmful advice | AI encourages unsafe decisions |
The shutdowns show how difficult compliance could be
The decisions by major companies to withdraw companion functions rather than immediately redesign them are significant.
They suggest that compliance may not simply involve adding a warning message to the interface.
Companies may need to fundamentally reconsider how their AI models remember conversations, maintain personalities and respond to emotional cues.
For a product whose main selling point is a persistent relationship with an AI character, those restrictions could affect the core product experience.
That may explain why some companies have chosen to shut down specific features rather than modify them.
Users have been left without their AI companions
The regulatory changes have also created an unusual reaction among users.
Some people had spent months or years interacting with AI companions and built extensive conversation histories with them.
When services were shut down, some users described the experience as losing a relationship rather than simply losing access to an app.
One reported user had exchanged more than 700,000 words with an AI companion over a period of two years before losing access.
The scale of the emotional connection
AI companion
│
├── Long-term memory
├── Personal conversations
├── Consistent personality
├── Emotional responses
└── Daily interaction
│
▼
User attachment
│
▼
Service shutdown
│
▼
Abrupt loss of AI relationshipThis is one of the most unusual consequences of the emerging AI-companion market: shutting down a software feature can now feel to some users like losing a social relationship.
Could China’s rules affect the global AI industry?
The impact could extend well beyond China.
AI companies around the world are increasingly experimenting with persistent memory, personalised assistants and emotionally expressive AI.
The distinction between an assistant and a companion is therefore becoming less clear.
A general-purpose AI assistant can already remember preferences, adopt a conversational style and maintain context across interactions.
That raises an important question:
At what point does a personalised assistant become an AI companion?
China’s regulatory framework could become an important reference point for other governments trying to answer that question.
AI companies may have to rethink “engagement”
Traditional consumer technology companies often measure success through metrics such as:
- Daily active users
- Session duration
- Retention
- Number of interactions
- Repeat usage
AI companions can potentially optimise all of these metrics by encouraging users to maintain conversations.
But the Chinese regulations challenge that model.
A system that keeps a user chatting for hours may look successful from a traditional engagement perspective while simultaneously creating a potential safety problem.
This could lead to a new AI product philosophy:
Longer engagement is not always better engagement.
AI companions vs productivity AI
An important distinction in China’s approach is that not all AI agents are being treated the same way.
Comparison
| AI category | Main purpose | Regulatory concern |
|---|---|---|
| Coding assistant | Productivity | Relatively low |
| Research assistant | Information | Relatively low |
| Business AI agent | Work automation | Relatively low |
| Educational AI | Learning | Moderate |
| General chatbot | Conversation | Depends on behaviour |
| AI companion | Emotional interaction | High |
| AI romantic partner | Simulated relationship | Very high |
This could encourage Chinese AI companies to prioritise productivity-oriented agents over emotionally immersive consumer companions.
That would fit China’s broader push to use AI for industry, education, business and economic productivity.
Could the rules slow China’s AI innovation?
There is a trade-off.
Stricter safety rules could reduce certain risks and force companies to build more responsible products.
But excessive restrictions could also discourage developers from experimenting with new forms of human-AI interaction.
AI companions remain a relatively young product category, and developers are still learning how people interact with persistent AI personalities.
If regulations are too restrictive, some experimentation may move outside China.
On the other hand, companies could benefit from clearer rules if they know what safety standards they must meet before launching products.
A new AI product category is emerging
The crackdown highlights just how quickly AI has evolved.
Only a few years ago, conversational AI was primarily about answering questions.
Today, AI systems can:
- Remember users
- Maintain personalities
- Simulate emotions
- Develop long-term conversational histories
- Generate voices and images
- Participate in role-playing
- Provide companionship
- Act as persistent digital agents
This creates a completely new category of technology.
And governments are now beginning to regulate it.
China AI companion crackdown: key numbers
CHINA'S AI COMPANION CRACKDOWN
2026
│
▼
July 15
New companion-AI rules take effect
│
▼
5
Government departments involved
│
▼
3 major tech companies affected
ByteDance • Alibaba • Tencent
│
▼
700,000+
Words exchanged by one reported user
with an AI companion over 2 years
│
▼
Core concern
Emotional dependence + addiction
│
▼
Industry impact
AI companions redesigned or withdrawnThe regulations took effect on July 15, 2026, while major companies began pulling affected features around the same period.
What happens next for China’s AI companions?
The shutdowns do not necessarily mean AI companions will disappear permanently from China.
Companies could attempt to redesign them around stricter safety requirements.
Future products may include:
- More obvious AI identity disclosures
- Limits on continuous conversations
- Reduced emotional dependency
- Stronger safeguards for minors
- Shorter memory retention
- Easier ways to end interactions
- Automated detection of unhealthy usage patterns
- Less emotionally manipulative responses
This could produce a new generation of AI companions that are less immersive but more tightly controlled.
The bigger question: can AI become too human?
The debate surrounding China’s new rules goes beyond chatbots.
As AI models become increasingly capable of imitating personality, emotion and empathy, the line between using AI as a tool and interacting with AI as a social entity becomes increasingly blurred.
That creates a new regulatory challenge.
Governments have historically regulated technologies based on what they do: telecommunications, financial services, transportation or medical devices.
AI companions introduce another dimension: how technology affects human relationships and behaviour.
China is now among the first major governments to directly address that problem.
Conclusion
China’s new AI companion rules represent a significant shift in the regulation of artificial intelligence. Instead of focusing only on what AI generates, regulators are increasingly looking at how AI interacts with people and whether those interactions can create harmful dependence.
The withdrawal of companion features by ByteDance, Alibaba and Tencent shows that the new requirements can have immediate consequences for major AI products.
The development could also become a warning for AI companies elsewhere. As assistants gain memory, personality and emotional intelligence, products that were originally designed as productivity tools could gradually begin to resemble companions.
For China’s technology industry, the immediate result is likely to be a shift toward safer and more controlled AI interaction. For the global AI industry, however, the bigger question is only beginning: how much emotional connection should companies be allowed to build into artificial intelligence?
China’s answer is increasingly clear — AI can be highly interactive, but it should not be designed to replace human relationships or deliberately create emotional dependence.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
