China’s top internet regulator has published a draft of broad new rules that would sharply limit how artificial intelligence can interact with children and teenagers, reflecting Beijing’s intensified effort to shape the architecture of online experiences in the domestic AI market. The provisions, released for public comment on September 18, 2026, are titled "Provisions of the State Council on Ensuring the Healthy and Safe Use of the Internet by Minors" and cover social networks, games, short video platforms and AI services. Among the most consequential proposals is an explicit ban on offering "virtual relatives or companions" and on services that could induce addiction or otherwise seriously affect minors' physical or mental health. The draft would require platforms to implement robust minors mode features for any online service that may affect the cognition of users under 16. That obligation spans conventional online products and increasingly sophisticated AI offerings, including chatbots and personalized generative services that advise or emotionally engage users. The package also sets out age tiers and access rules. It would limit social networking between strangers for younger users, bar access entirely for very young children to certain high risk services, and expand duties on platforms to strengthen age verification, parental controls, content governance and algorithm oversight. Public feedback on the draft is solicited through a comment window that closes on October 17, 2026. Why the draft matters for AI development and deployment China has already moved quickly in recent years to regulate generative AI and humanlike interaction. But this draft flags a decisive regulatory turn: the state is explicitly bringing AI companion services into the same regulatory frame as games and social apps that have long been targeted with youth protection rules. For Chinese AI companies that build conversational agents, digital companions or emotionally responsive applications, the draft signals immediate product and compliance consequences. First, companies may have to disable or significantly alter companion features for underage users, or lock access entirely behind rigorous age checks and minors mode defaults. That will change design tradeoffs for products that rely on long term user engagement, personalization or revenue streams tied to in app purchases. Second, the draft foregrounds algorithm governance. Platforms would need to demonstrate that recommendation logic, personalization signals and agent behaviors do not deliberately promote addiction, excessive consumption or unhealthy emotional dependence among minors. That will push teams to bake safety constraints, monitoring and audit trails into model development and deployment pipelines. Third, the rules extend data and privacy obligations. Age assurance systems and minors mode functions require more intrusive verification mechanics and reliance on behavioral signals, which in turn intersect with existing Chinese data protection and cybersecurity laws. Companies will need to navigate compliance tradeoffs between verifying user age and avoiding excessive collection of personal information. Regulatory context and enforcement levers Beijing’s draft sits on top of a growing stack of AI governance tools. National requirements for labeling AI generated content took effect in 2025, interim measures for anthropomorphic AI interaction services came into force in 2026, and multiple ministries have issued technical standards and measurement guidelines this year. Regulators have already required filing and registration for generative services and periodically removed unlawful AI generated content from platforms. The minors draft expands enforcement levers beyond content takedowns. It contemplates administrative supervision of platforms’ internal controls, mandatory technical measures such as minors mode switch defaults, and penalties for violations that could range from fines to restrictions on services. In practice, the Cyberspace Administration of China coordinates enforcement with other agencies that oversee market competition, public security and consumer protection, creating a cross agency compliance environment. Industry and international implications Large Chinese technology groups that operate consumer facing AI services will be most immediately affected. They already maintain substantial trust and safety teams and product governance functions; the draft will likely accelerate engineering work to isolate age segmented experience layers, harden moderation pipelines and document algorithmic decision making. Smaller startups and foreign firms face different challenges. Startups may find the compliance cost of segregating minors experiences and maintaining robust age assurance infrastructure difficult to absorb. Foreign companies offering AI services in or into China confront an existing reality of local filing requirements, data localization and restricted market access for some Western models. The draft may widen the gap between domestic and foreign providers by raising the bar for lawful operations aimed at minors. Globally, Beijing’s move adds to a mosaic of regulatory approaches to adolescent safety and AI. In contrast to more market permissive frameworks in some other jurisdictions, China is pushing function specific restrictions that limit both service design and the permissible behaviors of deployed models when interacting with children. For multinational companies, divergent rules for minors could mean multiple product builds and carefully partitioned model behaviors across jurisdictions. What to watch next The public comment period closes on October 17, 2026, a date that will matter to companies seeking to shape the final language of the provisions. Observers should watch whether Beijing narrows the scope of prohibited services, clarifies enforcement thresholds for algorithmic harms, or prescribes specific technical standards for age assurance. Regulators may also issue companion technical guidance or national standards that translate high level prohibitions into concrete engineering requirements. For developers building conversational AI and companionship features, the draft is a clear signal to prioritize safety by design, document risk assessments and accelerate work on age dependent controls. For parents, educators and policymakers, it confirms that China intends to make protective architecture an obligatory element of the country’s fast evolving AI ecosystem. The draft amplifies a central theme in China’s AI governance agenda: shaping not only what models can say, but how they can form relationships with users, especially the youngest ones. The outcome of this consultation will determine how that theme is encoded into the products used by tens of millions of Chinese minors in the years to come.