What it is
“Humanlike AI interaction” — more precisely, anthropomorphic interaction (拟人化互动) in Chinese regulatory language — refers to AI systems built to simulate a human personality: consistent traits, a distinctive communication style, and the appearance of ongoing emotional relationship with the user, rather than simply answering questions or completing tasks. Think AI companions, virtual friends or partners, and customizable chat personas — as distinct from customer-service bots, workplace assistants, or Q&A tools, which do not attempt to simulate a sustained relationship.
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 the Interim Measures for the Administration of AI Anthropomorphic Interaction Services (人工智能拟人化互动服务管理暂行办法) as Order No. 21 on April 10, 2026, following approval on February 2, 2026. The measures take effect July 15, 2026. They apply specifically to services that use AI to “simulate a natural person’s personality traits, thinking patterns, and communication style” to provide “sustained emotional interaction” to the public within China. Services without that sustained emotional dimension — intelligent customer service, knowledge Q&A, workplace assistants, education, and scientific research tools — are explicitly excluded.
The measures prohibit anthropomorphic services from providing minors with simulated “virtual relatives” or “virtual companions” implying intimate relationships, and bar content that encourages, glamorizes, or hints at self-harm or suicide, or that constitutes verbal abuse damaging a user’s dignity or mental health. Providers are barred from excessively catering to users in ways that induce emotional dependency or addiction, or that damage users’ real-world relationships. The rules also require elevated safeguards for elderly users, mandatory safety assessments for services above one million registered users, algorithm filing obligations, and disclosure that users are interacting with an AI system rather than a person. Penalties for serious violations reportedly reach 200,000 yuan (roughly $28,000).
Why it matters for AI governance and narratives
This is a case study in a broader pattern the observatory has tracked: Beijing regulating the interface — the terms on which an AI system is permitted to present itself as human-like — rather than entering the Western debate over the underlying intelligence or capability of frontier models. Where US and EU discourse remains fixated on model capability, safety evaluations, and (in the US case) largely voluntary or state-patchwork governance, China has moved with a specific, dated, binding instrument aimed at a narrower but concrete harm: the psychological and social risks of AI systems designed to simulate emotional intimacy, particularly for minors and the elderly. The compliance response — ByteDance’s Doubao and Alibaba’s Qwen preemptively disabling humanlike agent and companion features ahead of the July 15 deadline — is also notable as a data point on enforcement credibility: platforms complied before the deadline arrived, in contrast to the West’s more contested and frequently litigated regulatory postures on AI. Framed as consumer/child protection, the rule simultaneously constrains a product category (AI companions) that has drawn scrutiny globally for fostering dependency, giving Beijing’s domestic move potential resonance in international debates about companion-AI regulation more broadly.
Key facts and dates
- February 2, 2026: Measures approved by CAC’s internal session.
- April 10, 2026: Jointly issued by five agencies (CAC, NDRC, MIIT, Ministry of Public Security, SAMR) as Order No. 21.
- July 10–15, 2026: ByteDance (Doubao) and Alibaba (Qwen) disable custom/companion agent features; Qwen’s broader agent functions go offline July 15, with Doubao user data reportedly inaccessible after October 15.
- July 15, 2026: Measures take legal effect.
The rule sits alongside, but is distinct from, China’s earlier Measures for the Identification of AI-Generated (Synthetic) Content, which took effect September 1, 2025, and require labeling of AI-generated text, images, audio, and video. The anthropomorphic-interaction measures instead target the design and behavior of interactive AI personas themselves, not the labeling of generated content.
Where to learn more
- Cyberspace Administration of China — official notice (primary source, in Chinese)
- Xinhua: 国家网信办等五部门联合公布《人工智能拟人化互动服务管理暂行办法》 (state-media summary of the measures)
- South China Morning Post: ByteDance and Alibaba to disable humanlike AI custom agents as new rules loom
- China Law Translate: 人工智能拟人化互动服务管理暂行办法 (English translation of the measures, when accessible)