China's AI companion chatbot rules take effect, forcing shutdown of user-built personas

An analysis on Just Security examines China's first AI companion rule, which took effect on 15 July 2026. On that day, according to the piece, China's two largest AI platforms closed down the customised AI persona services that millions of users had built, ten days after a notice ordering them to stop. The author contrasts this with the United States, where regulators have noticed the same risk of emotional dependency but responded differently.

The rule came out on 10 April 2026 from the Cyberspace Administration of China, working with four other agencies. It covers services imitating a person's personality and communication style that sustain emotional interaction, and exempts ordinary customer service, question-and-answer, work and research tools. Providers may not design services to cause emotional dependence or to replace real relationships. Minors cannot be offered virtual partners or relatives, and users under 14 need parental consent for other such services. Providers must tell users they are speaking to AI, show a break reminder after two hours of continuous use, and intervene in a crisis. Safety assessments are required at launch, after major changes, and at one million registered or 100,000 monthly active users.

Reactions included mourning for personas modelled on deceased relatives and relief among some parents. The article notes that the meaning of sustained emotional interaction is unclear, that dependence is not scored separately in the assessment, and that rollout was uneven. Doubao and Qwen removed custom personas, while ByteDance redirected users to its Maoxiang app, whose policy describes weaker age checks than announced.

The author argues that US state laws focus on suicide prevention and minors, that no federal rule exists for adults, and that the FTC's September 2025 inquiry has produced no rule or enforcement.