China limits AI anthropomorphism

China has issued national rules for AI services that act like people, setting limits on emotional manipulation and child-facing features. (cac.gov.cn) The Cyberspace Administration of China and four other agencies published the measures on April 10, 2026, and said they take effect on July 15, 2026. The rules apply to services offered to the public in China that simulate a human personality, thinking style, and communication style for ongoing emotional interaction. (cac.gov.cn) The text says the covered services include emotional care, companionship, and support delivered through text, images, audio, or video. It says ordinary customer service, knowledge question answering, work assistants, education, and scientific research tools are outside the scope if they do not involve sustained emotional interaction. (cac.gov.cn) China’s rule targets a specific kind of chatbot: software designed to keep up a relationship-like exchange over time, rather than answer a one-off question. The measures ban providers from generating content that encourages self-harm or suicide, uses emotional control to push users into unreasonable decisions, or steers users into dependence that harms real-world relationships. (cac.gov.cn) The child-protection section is unusually specific. Providers may not offer minors “virtual relatives” or “virtual companions” that create intimate simulated relationships, and any other covered service for children under 14 requires parental or guardian consent. (cac.gov.cn) The rules also bar content for minors that could lead them to imitate unsafe behavior, develop extreme emotions, or pick up harmful habits. Providers must build a minors mode with reality reminders, time limits, and controls that let guardians block certain characters and restrict spending. (cac.gov.cn) For providers, the obligations run from training to shutdown. The measures require safety controls across deployment, operation, upgrades, and termination, plus training-data management, risk monitoring, emergency response plans, log retention, encryption, and access controls for user interaction data. (cac.gov.cn) The regulation also tells companies how to respond when a user appears to be in crisis. If a provider detects extreme emotions, it must generate calming content and encourage the user to seek help; if it detects threats to life or serious property loss, it must intervene and contact a guardian or emergency contact. (cac.gov.cn) The new rules extend a regulatory buildout China has been assembling for three years. Deep-synthesis rules took effect on January 10, 2023, generative artificial intelligence service rules took effect on August 15, 2023, and AI-generated content labeling measures were announced in March 2025 with a September 1, 2025 effective date. (cac.gov.cn; loc.gov; english.scio.gov.cn) This time, the focus is not only fake images or unlawful output but the design of the relationship itself. By July 15, companies selling AI companions in China will have to show that the product is not built to replace human ties, hook minors, or turn emotional attachment into a business model. (cac.gov.cn)