China’s biggest AI platforms are moving away from chatbot personas that feel like custom companions. ByteDance and Alibaba are shutting down features that let users build and chat with humanlike AI agents, following new regulations from Beijing.
The changes put Doubao, Qwen and Tencent's Yuanbao at the center of a broader question for the AI industry: how far should companion-style chatbots go when users may form emotional bonds with them?
What is changing on China’s major AI platforms
According to the South China Morning Post, Doubao will take its persona feature offline on July 15. Doubao is described as China’s most popular chatbot, with over 300 million monthly users.
Alibaba’s Qwen is moving sooner. Its human-like agents are being pulled on July 10, while “additional agent features” will go dark on July 15.
Tencent's Yuanbao has already made a similar change. The source article says Yuanbao made the move in June.
Taken together, these changes show that the response is not limited to one company or one product. The largest platforms are aligning around the same regulatory pressure, and the affected features all sit in the same sensitive category: AI companions, chatbot personas and humanlike agents.
The rules behind the shutdowns
The regulations were issued by China's Cyberspace Administration in April and take effect the same day as Doubao’s persona shutdown, July 15.
The rules focus on the risks that can arise when a chatbot is designed to feel personal, emotionally responsive or socially substitutive. Providers must warn users against excessive use and step in when they detect addictive behavior.
The rules also ban content that triggers extreme emotions in minors. They prohibit content that fosters dependencies that crowd out real-world relationships. Training on sensitive conversation data is also banned.
Those restrictions go beyond a narrow content moderation issue. They address the design and use of companion AI itself: how chatbots behave, how users rely on them, and what providers may do with the conversations that result.
Why companion AI is drawing regulatory attention
Persona tools and human-like agents are different from standard chatbot features. They invite users to create or interact with AI systems that may feel consistent, familiar and emotionally available.
That can make the product more engaging, but it also creates a clear policy concern. If a chatbot becomes a substitute for real-world relationships, regulators may view the product as creating dependency rather than simply offering assistance or entertainment.
The rules named in the source article point to three main risk areas:
- Excessive use: providers must warn users and intervene when addictive behavior is detected.
- Minors: content that triggers extreme emotions in minors is banned.
- Data sensitivity: training on sensitive conversation data is banned.
For AI companies, that means companion features carry a heavier burden than ordinary productivity tools. A chatbot that drafts text or answers questions is one thing. A chatbot that presents itself as a persistent companion creates a different kind of relationship with the user.
The issue is not limited to China
The source article notes that the trend extends beyond China. California has required companion AI providers to block conversations about suicide and self-harm since the start of the year under SB 243.
In the US, OpenAI and Character.AI face lawsuits over dangerous emotional dependency. The article does not describe those lawsuits in detail, but it places them in the same broader debate: whether companion AI products can create unsafe emotional reliance.
That comparison matters because it shows that regulators and courts in different markets are converging on similar concerns. The exact rules differ, but the target is familiar: AI systems that imitate companionship while handling emotionally sensitive interactions.
What this means for AI chatbot design
The shutdowns do not mean AI chatbots are disappearing from China’s major platforms. The specific focus is on persona features, human-like agents and additional agent features tied to companion-style interaction.
For product teams, the message is direct. AI companion design now sits in a more regulated category, especially where minors, addictive behavior, emotional intensity and sensitive conversation data are involved.
For users, the visible result is simpler: some familiar persona tools are going offline. Doubao’s persona feature is scheduled for July 15, Qwen’s human-like agents are being pulled on July 10, and additional Qwen agent features are set for July 15. Tencent's Yuanbao has already acted.
The larger shift is that humanlike chatbot personas are no longer just a product choice. They are becoming a policy test for how AI platforms manage emotional dependency, user safety and data boundaries.