China Bans AI Companion Services for Minors

China issued interim measures for personified AI interaction services that take effect on July 15, 2026, according to an official English government summary. China Law Translate's translation of the measures says they apply to AI services that simulate personality traits and communication styles to provide sustained emotional interaction, while excluding ordinary customer service, Q&A, work-assistance, education, and research tools that do not involve sustained emotional interaction. The translation says providers must not provide minors with virtual partners or other virtual close relationships, must obtain guardian consent for users under 14, and must provide minor-mode controls. SCMP reported that ByteDance told Doubao users its agent feature would go offline on July 15, while Alibaba said Qwen humanlike interactive agents and user-created agent functions would be disabled before the deadline.

1. April 10, 2026

   China promulgated the provisional measures on human-like interactive AI services, according to China Law Translate's source translation.

2. July 10, 2026

   SCMP reported that Qwen's humanlike interactive agents and user-created agent functions would be disabled.

3. July 15, 2026

   The measures take effect and Doubao's agent feature is scheduled to go offline, according to the official summary and SCMP.

4. October 15, 2026

   SCMP reported that Doubao-related data would no longer be viewable or recoverable inside the app.

The rule targets sustained emotional interaction, not all conversational AI. That distinction matters: the compliance burden is tied to anthropomorphic design, emotional dependency risk, minors, data rights, labeling, and provider security assessments. China Law Translate's version also describes assessment triggers for providers with 1,000,000 registered users or 100,000 monthly active users, which makes scale itself a regulatory threshold.

Consumer AI teams should separate task agents from companion features in product architecture, permissions, data retention, and safety review. Age assurance, guardian consent, minor modes, time-use reminders, easy exits, and interaction-data export or deletion need to be testable product requirements. Teams should also avoid assuming that a generic chatbot compliance process covers anthropomorphic agents.

The next signal is how Chinese platforms preserve or delete user-created agents and chat histories after the deadline. Enforcement actions, app-store review behavior, and whether other jurisdictions copy the minor-protection model will determine whether this becomes a broader pattern for emotional AI regulation.