Key Takeaways
- Doubao, Qwen, and Yuanbao must disable persona features by July 15 under China's new AI companion rules
- Regulations target emotional dependency, addictive behavior, and parasocial relationships with AI
- Global AI companion regulation is accelerating, with California's SB 243 already in effect since January
China's three largest AI platforms are pulling the plug on humanlike chatbot personas. ByteDance's Doubao, Alibaba's Qwen, and Tencent's Yuanbao must disable features that let users create and interact with custom AI companions by July 15, following new rules from the Cyberspace Administration of China that take effect the same day.
Doubao, China's most popular AI chatbot with over 300 million monthly active users, will shut down its persona feature on July 15. Qwen moves faster, killing human-like agents on July 10 with additional agent capabilities going offline five days later. Tencent already made the shift in June, getting ahead of the deadline.
What the regulations actually require
The Cyberspace Administration issued these rules in April 2026. They mandate that AI providers warn users against excessive use and intervene when they detect addictive behavior. Content that triggers extreme emotions in minors or fosters dependencies that "crowd out real-world relationships" is banned outright. Training AI models on sensitive conversation data is also prohibited.
This effectively ends China's booming AI companion market. Over the past two years, Chinese platforms had differentiated themselves from Western competitors by leaning hard into persona creation. Users could build AI friends, romantic partners, and therapeutic companions. That market segment is now illegal.
Why China moved first
Beijing has been the most aggressive AI regulator in the world. The 2023 Generative AI regulations required algorithm registration and content moderation. The 2024 deepfake rules tightened controls on synthetic media. This latest action fits the pattern: act fast, force compliance from domestic tech giants, and frame the intervention around social stability and mental health.
The framing matters. China isn't saying AI companions are dangerous because of misinformation or fraud. The concern is psychological: parasocial relationships with AI, emotional manipulation, and harm to young users who might prefer chatbot friendships to human ones. That's a harder problem to quantify, but Chinese regulators don't need to prove measurable harm before acting.
The US is watching, and suing
China isn't alone in this concern. California's SB 243, which took effect in January, requires AI companion providers to block conversations about suicide and self-harm. It's narrower than China's rules but addresses the same underlying worry.
OpenAI and Character.AI currently face lawsuits in the US over alleged dangerous emotional dependency. Character.AI in particular has been scrutinized after reports of users forming intense attachments to its personas. The legal exposure is real, but American regulation remains reactive where China's is preemptive.
What this means for AI product teams
If you're building conversational AI that touches Chinese users, the persona playbook is closed. Any feature that encourages users to treat an AI as a human relationship is now a compliance risk in the world's largest AI market.
The harder question is whether Western regulation follows. SB 243 passed with broad support in California. The EU's AI Act includes provisions on transparency and emotional manipulation. The regulatory direction is clear even if the timeline isn't.
Product teams building companion-style features should assume that some version of these restrictions will arrive in their markets within 18 to 24 months. That means building guardrails now, not after the rule takes effect.
Logicity's Take
China just killed the stickiest feature category in consumer AI. Persona chatbots drove engagement numbers that utility bots couldn't match. ByteDance and Alibaba now have to compete purely on task completion, where the product differentiation is thinner and the switching costs are lower. For Western AI builders, this is a preview of regulatory pressure to come. If your roadmap includes AI companions, therapeutic bots, or any feature designed to create emotional attachment, start documenting your safety measures and intervention protocols now. The US lawsuit route is slower than Chinese regulation, but Character.AI's legal bills suggest the bill comes due eventually.
Frequently Asked Questions
Which Chinese AI chatbots are affected by the new regulation?
ByteDance's Doubao, Alibaba's Qwen, and Tencent's Yuanbao must all disable humanlike persona features by July 15, 2026. Yuanbao already complied in June.
What behaviors do China's AI companion rules prohibit?
The rules ban content that triggers extreme emotions in minors, fosters dependencies that replace real-world relationships, and enables addictive chatbot use. Providers must warn against excessive use and intervene when they detect addictive behavior.
How does California's SB 243 compare to China's AI companion rules?
SB 243 is narrower. It requires AI companion providers to block conversations about suicide and self-harm, but doesn't ban persona features entirely or target emotional dependency as broadly as China's rules do.
Can users still use Doubao after July 15?
Yes, but without the persona feature. Users can still interact with Doubao for standard AI assistant tasks. Only the custom companion and human-like agent capabilities are being removed.
Are Western AI companies facing similar restrictions?
Not yet at the regulatory level, but OpenAI and Character.AI face lawsuits over emotional dependency concerns. California's SB 243 addresses a subset of the same issues. More regulation is likely coming.
Another example of AI capability outpacing governance frameworks
Need Help Implementing This?
If you're building conversational AI and need to implement safety guardrails before regulations arrive in your market, our team at Logicity tracks global AI policy developments and can help you design compliant user experiences. Get in touch through our consulting page.
Source: The Decoder / Jonathan Kemper
Manaal Khan
Tech & Innovation Writer
Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.
