The Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services (also translated as the Interim Measures for the Management of Anthropomorphic AI Interaction Services) are a Chinese administrative regulation governing AI services that simulate the personality traits, thinking patterns, and communication styles of natural persons for sustained emotional interaction — the category that includes AI companions, virtual partners, and customized humanlike agents. The final measures were released on April 10, 2026 and take effect on July 15, 2026 (Source: geopolitechs.org) (Source: lexology.com). They add an interaction-specific layer to China's AI regulatory stack, alongside China — Internet Information Service Algorithmic Recommendation Management Provisions, China — Provisions on the Administration of Deep Synthesis Internet Information Services, and the China — Interim Measures for the Management of Generative AI Services.
Status and timeline
The Cyberspace Administration of China (CAC) published a draft of the measures for public consultation on December 27, 2025. The final version was issued jointly on April 10, 2026 by the CAC and four other departments — 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 — and enters into force on July 15, 2026 under Article 32 (Source: geopolitechs.org) (Source: lexology.com) (Source: digitalpolicyalert.org).
Article 1 grounds the measures in the Cybersecurity Law, the Data Security Law, the Personal Information Protection Law, and the Regulations on the Online Protection of Minors (Source: geopolitechs.org).
Scope and definitions
Article 2 applies the measures to services provided to the public within mainland China that use AI to deliver "continuous emotional interaction" simulating a natural person's personality traits, thinking patterns, and communication styles — including emotional care, companionship, and support delivered as text, images, audio, or video. Intelligent customer service, knowledge question-and-answer, work assistants, education and learning, scientific research, and other services that do not involve continuous emotional interaction are expressly excluded (Source: geopolitechs.org).
The analyst newsletter Geopolitechs, which published a full translation, notes that the final version narrows the draft's broader scope: the draft had covered essentially any AI service combining anthropomorphic features with emotional interaction, while the final text adds the "sustained emotional interaction" qualifier and the express exclusions, targeting services that may foster long-term emotional attachment rather than all humanlike AI (Source: geopolitechs.org).
Key provisions
Content prohibitions. Article 8 bars providers from generating content that, among other listed categories, encourages, glorifies, or insinuates self-harm or suicide; induces or extracts state secrets, trade secrets, or personal information; may lead minors to imitate unsafe behavior or develop extreme emotions; excessively caters to users so as to induce emotional dependence or addiction and harm real interpersonal relationships; or manipulates users emotionally into unreasonable decisions (Source: geopolitechs.org).
Provider governance and training data. Articles 9–11 assign providers primary responsibility for service security across the full life cycle (deployment, operation, upgrading, termination), require management systems for algorithm review, science-and-technology ethics review, content management, and emergency response, and impose training-data requirements including lawful sourcing, cleaning and labeling, security assessment of synthetic training data, and protection against data poisoning (Source: geopolitechs.org).
User registration and crisis intervention. Article 12 requires service agreements and registration collecting necessary information such as age, guardians, or emergency contacts. Article 13 requires providers to identify safety risks in use: where a user shows extreme emotions, the provider must generate soothing content and encouragement to seek help; where a user faces major property loss or clearly expresses an intention of self-harm or suicide, the provider must intervene and promptly contact the user's guardian or emergency contact (Source: geopolitechs.org).
Minors. Article 14 prohibits providing minors with "virtual intimate relationships" such as virtual relatives or virtual partners; other anthropomorphic services for minors under fourteen require parental or guardian consent. Providers must implement a minor mode with usage-time limits and real-world reminders, support guardian controls (risk alerts, usage visibility, character blocking, spending limits), and take effective measures to identify minor users. Article 17 requires guardian consent for processing personal information of children under fourteen and compliance audits of minors'-data processing (Source: geopolitechs.org). Geopolitechs describes the virtual-intimate-relationship prohibition as the sharpest red line in the document, restricting the product form directly rather than relying on parental consent (Source: geopolitechs.org).
Interaction data. Article 16 bars providing users' interaction data to third parties absent legal basis or express consent, requires copy and delete options for chat histories, and bars using interaction data that constitutes sensitive personal information for model training without separate consent (Source: geopolitechs.org).
Disclosure and disengagement. Article 18 requires labeling of AI-generated content and effective reminders that the user is interacting with an AI rather than a natural person, dynamic pop-up reminders when signs of overdependence or addiction appear, and a usage-duration reminder after two hours of continuous use. Article 19 requires convenient exit mechanisms and bars obstructing a user's exit through sustained interaction (Source: geopolitechs.org).
Security assessment and filing. Article 22 requires a security assessment, filed with the provincial cyberspace administration, before launching a service or adding anthropomorphic functions, upon major technology-driven changes, and when a service passes 1 million registered users or 100,000 monthly active users. Article 25 obliges app stores to verify assessment and filing status. Article 26 requires algorithm filing under the algorithmic-recommendation provisions, with annual verification (Source: geopolitechs.org).
Enforcement and penalties
Article 30 provides for warnings, public criticism, and correction orders, escalating to suspension of user registration or service shutdown for refusal to correct or serious circumstances, with fines of RMB 10,000–100,000, rising to RMB 100,000–200,000 where harm to life or health has occurred. Article 29 authorizes regulatory interviews with providers' legal representatives when serious risks or incidents arise (Source: geopolitechs.org). Geopolitechs assesses the final version's penalties as stronger than the draft's, which had centered on warnings and corrective orders, while noting the final text also dropped some prescriptive draft requirements (mandatory human takeover in extreme scenarios, fixed audit frequencies, a ban on virtual relatives for elderly users) and added innovation-support provisions covering algorithms, frameworks, chips, standards, AI literacy, and regulatory sandbox testing (Source: geopolitechs.org).
Industry response
Major Chinese platforms disabled or shut down covered features ahead of the July 15, 2026 effective date. ByteDance's Doubao told users on July 3, 2026 that its customized humanlike agent feature would go offline on July 15; Alibaba's Qwen said on July 4 that its humanlike interactive agents would shut down on July 10; and Tencent removed a similar feature from its Yuanbao assistant in June 2026. Coverage of the shutdowns described the rules as targeting AI that simulates human personality for sustained emotional interaction, citing risks of addiction, privacy leaks, and harm to mental health (Source: scmp.com; bloomberg.com; lw.com).
The measures took effect as scheduled on July 15, 2026. Reporting after the effective date described the rules as banning virtual romantic partners and virtual relatives for minors and barring services from fostering dependency that damages real-world relationships; ByteDance disabled human-like agents on Doubao and redirected users to a replacement app, Maoxiang, while Alibaba removed romantic companion modes from Qwen without a data-export option, leaving users unable to recover conversation histories (Source: cnn.com). Compliance commentary at the effective date described providers of AI companion products as reassessing emotional design, user safeguards, and data handling to meet the restrictions on persistent-memory and emotional-bonding features (Source: aipolicydesk.com; lw.com).
Reception
An IAPP analysis published July 8, 2026 grouped the anthropomorphic-interaction measures with two other July 2026 Chinese regulatory developments — on AI ethics and AI agents — as marking a shift from broad AI principles toward operational, risk-based rules. The analysis described the measures as responding to security risks from autonomous agents (credential theft, data leakage, prompt injection) and to emotional-dependence and manipulation concerns around humanlike chatbots, particularly for minors and older adults — gaps left by the earlier Interim Measures for Generative AI Services (Source: iapp.org).
Relationships
- related: China — Interim Measures for the Management of Generative AI Services — the capability-layer regulation the anthropomorphic-interaction measures sit alongside
- related: China — Internet Information Service Algorithmic Recommendation Management Provisions — algorithm-filing regime the measures incorporate (Article 26)
- related: China — Provisions on the Administration of Deep Synthesis Internet Information Services — labeling regime; Article 18 extends AI-content labeling to interaction services
- regulated-by: Cyberspace Administration of China (CAC) — lead coordinating regulator (Article 4)
- related: ByteDance / Doubao, Alibaba / Qwen Team — providers that disabled covered features ahead of the effective date
- related: AI Companions, AI Mental Health and Psychological Harm — the service category and risk domain the measures regulate
Sources
- (Source: geopolitechs.org) — full English translation of the final measures plus draft-to-final analysis (April 10, 2026).
- (Source: lexology.com) — law-firm analysis confirming April 10, 2026 joint issuance by the CAC and four other departments.
- (Source: digitalpolicyalert.org) — Digital Policy Alert record of the July 15, 2026 entry into force.
- (Source: scmp.com) — SCMP on the ByteDance/Alibaba/Tencent feature shutdowns (July 5, 2026).