China enforces new AI companion rules as Doubao and Qwen shut agents

China’s new rules on anthropomorphic AI interaction services are one of the clearest attempts yet to regulate not just AI content, but the *relationship* between humans and AI systems. By targeting virtual companions, custom “agents” and other emotionally engaging chatbots, regulators are drawing a bright line around products that might induce dependency, blur the boundary between humans and machines, or expose minors to psychological risk. For ByteDance and Alibaba, being forced to turn off Doubao and Qwen-based agents is a concrete product hit, and a signal that China will tolerate frontier models only within tightly defined behavioral and UX constraints.([finance.sina.com.cn](https://finance.sina.com.cn/stock/t/2026-07-15/doc-inihweze8070643.shtml))

For the race to AGI, this is less about slowing core model research and more about constraining commercialization paths. Labs can still train large models, but any offering that feels too human, persistent, or intimate will now carry heavy compliance overhead. That tends to favor enterprise and productivity use cases over mass-market AI companions. It may also push Chinese firms to innovate in explainability and safety tooling—monitoring emotional interactions, setting use-time limits, and segmenting minors—because these capabilities are now regulatory requirements, not optional extras. Over time, similar rules could become a template for other jurisdictions wrestling with AI “pseudo-relationships.”