China's rules on anthropomorphic AI companion services take effect

A report says that China's interim measures on anthropomorphic AI interactive services, which run to 32 articles, took effect on 15 July 2026. They were issued on 10 April 2026 jointly by the Cyberspace Administration of China and four other bodies, among them the NDRC and MIIT.

The measures cover services that imitate the personality, thinking and communication style of real people to give the public ongoing emotional interaction, such as companionship or virtual characters. Tools for customer service, Q&A, work assistance or education are excluded unless they involve sustained emotional interaction.

Providers must tell users they are dealing with AI, show prominent reminders when users appear over-reliant, and prompt users after two hours of continuous use. Minors may not be offered virtual partner or relative services, other services for under-14s need parental or guardian consent, and a minors' mode must include time limits, spending controls and guardian alerts. Older users must get guidance and quick answers to requests for help. Providers may not aim to replace social interaction, induce dependence or manipulate emotions, and must step in when users express an intention to self-harm. Interaction data must be encrypted, sensitive information needs separate consent before training use, and users must be able to delete their chat history and leave easily.

Safety assessments filed with provincial cyberspace authorities are required at launch, after significant changes, at one million registered or 100,000 monthly active users, and where national security or the public interest may be affected. Algorithm filing is also required. Penalties run from warnings to service suspension and fines of RMB 10,000 to 100,000, rising to RMB 200,000 where life or health is harmed.