On April 10, 2026, China’s Cyberspace Administration (CAC), together with the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration for Market Regulation, issued Order No. 21: the Provisional Measures on the Administration of Human-like Interactive Artificial Intelligence Services, effective July 15, 2026. It is the second time China has written standalone rules for a specific AI service category, after the 2023 generative AI measures — and this time the target is AI companions that simulate a real person’s personality and emotional interaction.
The measures turn “anti-addiction” and “anti-manipulation” from product ethics into enforceable legal duties. Compared with the disclosure-style chatbot laws emerging elsewhere, the Chinese version goes further: it regulates the interaction design itself.
What Falls in Scope: A Precise Definition
Article 2 draws the boundary: any service that simulates a natural person’s personality, thinking, and communication style to provide “sustained emotional interaction” — emotional care, companionship, support — is covered. Ordinary smart customer service, informational Q&A, work assistance, and education or research tools without that interaction are explicitly excluded.
The notable part is how the line is drawn: not by technology (whether you use an LLM), but by interaction form (whether you cultivate a sustained emotional relationship). Swapping your model provider does not exempt you; the product shape is what counts.
Anti-Addiction and Anti-Manipulation as Statutory Duties
Article 8 bans content that “excessively panders” to users or induces emotional reliance and addiction, and bars using “emotional manipulation” to push users toward unreasonable decisions. Article 10 is blunter still: service goals must not include “replacing social interactions, psychologically controlling users, or inducing addiction.”
Article 18 pushes the duty down to product details: AI-generated content must be labeled; users must be told they are talking to an AI, not a human; pop-up warnings are required when overreliance or addiction tendencies appear; and a reminder must fire after every two hours of continuous use. These are concrete, testable, auditable requirements.
Minors and Crisis Intervention
Minor protection is the hardest block. Virtual partners and “virtual close relationships” are banned outright for minors; users under 14 require guardian consent and are forced into a minors mode with usage limits, periodic real-world reminders, character blocking, and charge restrictions. Handling of minors’ data is subject to annual audits.
Article 13 adds a crisis-response duty: when extreme emotions are detected — self-harm or suicidal tendencies, or a major personal loss — the service must generate consoling content and contact the user’s guardian or emergency contact. In effect, “emotion detection plus human notification” becomes a platform obligation.
Data, Exit Rights, and the Million-User Threshold
On data: interaction data cannot go to third parties without consent; users can reproduce or delete it; sensitive personal information cannot train models absent independent consent. Article 19 introduces an unusual “exit right”: providers must offer convenient exit channels and may not impede departure through continued interaction — a direct strike at retention-first companion design.
Scale thresholds are codified too: security assessments are triggered by service launches, new technology, one million or more registered users or 100,000 or more monthly actives, or national security risks. Penalties range from warnings and ordered corrections to service suspension; refusal to correct draws fines of 10,000–100,000 RMB, rising to 100,000–200,000 RMB where life- or health-harming consequences result.
What It Means for Product Teams
There are exactly three months between promulgation and effect. Companion products serving the Chinese market need, before July 15: registration and age verification, collection of guardian and emergency-contact details, continuous-use timers with reminders, addiction-tendency detection, crisis escalation flows, and redesigned offboarding. Each is real engineering work, not a compliance statement.
The bigger picture: app stores get gatekeeping duties (Article 25), algorithm filing with annual checks becomes routine (Article 26), and the state will stand up AI sandbox testing platforms (Article 28). Set against Washington State’s first AI chatbot safety law (see Washington’s AI chatbot safety law), “disclose the AI, protect minors” is now a cross-jurisdiction baseline; the difference is that China elevates anti-addiction and anti-manipulation to penalty level, reaching directly into interaction design.
Sources
- 人工智能拟人化互动服务管理暂行办法 (official text) — CAC
- Provisional Measures on Human-like Interactive AI Services — China Law Translate
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.