Summary

• China’s rules cover AI services that simulate human personality and communication styles for ongoing emotional interaction, while generally excluding ordinary work, education, research, and non-emotional customer-service tools.
• Providers must conduct safety assessments, file algorithms at defined launch or user thresholds, disclose that users are interacting with AI, detect emotional distress or dependence, and provide crisis-response and exit mechanisms.
• The measures impose stronger protections for minors, including age verification, guardian consent for younger users, minor modes, parental controls, and a ban on virtual intimate relationships for minors.
• Interaction data must be protected with encryption and access controls, and sensitive interaction data cannot be used for model training without separate consent or legal authorization.
• Reports describe companion-feature shutdowns or rollbacks by Doubao, Qwen, Tencent, and NetEase Cloud Music as the rules took effect.
• The requirements may conflict with product designs based on persistent memory, stable personas, and continuous emotional engagement.
• China’s targeted regime contrasts with the EU’s risk-based AI Act implementation and the voluntary U.S. pre-release evaluation framework described in the material.