The Cyberspace Administration of China has opened public comment on draft State Council provisions that would route AI services capable of affecting minors' cognition through minors mode for under-16 users and would prohibit algorithm models that induce emotional dependence, addiction or excessive consumption in minor users

A consultation on a draft, and nothing more yet

On 18 September 2026 the Cyberspace Administration of China posted a notice inviting public comment on a draft State Council regulation, 国务院关于保障未成年人健康安全使用网络的规定（征求意见稿）, which we translate as the State Council Provisions on Safeguarding Minors' Healthy and Safe Use of the Internet (draft for comment). The notice says CAC organised the drafting under the Minors Protection Law, the Cybersecurity Law and the Personal Information Protection Law. It binds nobody. Comments are accepted by email to weibao@cac.gov.cn or by post to the CAC Network Rule of Law Bureau at 北京市西城区车公庄大街11号, postcode 100044, with the envelope marked 未成年人健康安全使用网络征求意见, until 17 October 2026.

The notice page carries the full 25-article draft and a drafting explanation inline, so there is no separate attachment to chase. The explanation confirms the article count and says CAC drafted the text with relevant departments after research, consultation with central and local bodies, platform companies and academics, and a legislative assessment. It also names the convergence of the internet with artificial intelligence and big data as one reason the drafters give for the new rules, and says recent legislation abroad on minors' online safety was studied.

One line in the draft deserves its own sentence. Article 25 reads 本规定自2026年 月 日起施行, with the month and day blank. The year printed is a drafting placeholder; it is not a statement that the regulation will be adopted in 2026, and we do not read it as one. The draft is titled as a State Council instrument, but the notice comes from CAC and says nothing about State Council adoption. We describe this as a CAC consultation on draft State Council provisions, not as an act of the State Council.

Article 4 would prohibit; Article 5 would gate through minors mode

The two articles do different work and the difference matters for anyone building conversational products. Article 4, as drafted, would prohibit network service providers from offering listed services to minors at all. Item (2) is the one for AI teams: 虚拟亲属、虚拟伴侣等虚拟亲密关系服务, which we translate as virtual intimate relationship services such as virtual relatives and virtual partners. Item (1) would bar stranger social networking services, with an exception for minors aged 16 or above whose main source of living is their own labour income; that exception attaches to item (1) only. Item (3) would bar services that induce addiction or otherwise harm or may seriously affect minors' physical and mental health, and item (4) would bar livestream publishing to under-16s. Minors mode is not a route around Article 4: a prohibited service stays prohibited in any mode.

Article 5, also a proposal, takes a different approach for under-16s. Where a network service provider offers the listed services 向未满十六周岁的未成年人, to minors under sixteen, it would have to provide them through 未成年人模式, minors mode. Item (3) is 可能影响未成年人认知的人工智能服务: artificial intelligence services that may affect minors' cognition. The other items are audiovisual, non-publishing livestream and gaming services, social services including microblogs, forums, groups, moments, private messaging and comment threads, and a catch-all for other services with a significant effect on minors.

Two limits in the text keep Article 5(3) narrower than a plain reading of the headline. The trigger is a service that may affect minors' cognition, and the draft does not define that phrase or give a test for it. The mode requirement applies to provision to under-16s; for minors aged 16 and above, the second paragraph of Article 24 says minors-mode provision is encouraged, not required. The first paragraph of Article 24 is the one that makes the categories movable: 国家网信部门可以会同国务院有关部门根据保障未成年人健康安全使用网络的需要，对本规定第四条、第五条所列网络服务类型适时予以调整, which we translate as the national cyberspace department may, together with relevant State Council departments and as the need to safeguard minors' healthy and safe use of the internet requires, adjust in due course the types of network service listed in Articles 4 and 5. Whether a given chatbot, recommendation feed or education product falls inside item (3) is a question the text leaves open, and we do not answer it for any product.

Article 8: review and management systems, and a model prohibition

Article 8 is the algorithm provision and the source of our quotation. As drafted it would require network service providers to establish and improve systems for 算法机制机理审核 (algorithm mechanism review), 科技伦理审查 (science and technology ethics review), information content management, and network and data security management; to deploy content management technical measures and personnel suited to the type and scale of the service and the characteristics of its minor users; and, in its final clause, 不得设置诱导未成年人用户情感依赖、沉迷以及过度消费等的算法模型.

Our translation of that final clause: the draft would prohibit providers from setting up algorithm models that induce emotional dependence, addiction, excessive consumption or the like in minor users. The 不得 that opens the clause is the prohibitive formula of Chinese legislative drafting; it belongs to the proposed text, and nothing in it is in force. The 等 at the end makes the list illustrative, not closed. The clause does not say designed to induce, and it contains no intent element; we translate it as a prohibition on setting up models that induce the listed effects, and we do not add an intent test the text does not have.

What Article 8 does not do is as important as what it does. The review systems it names are the provider's own internal systems. The article does not create a government pre-approval step for models, and it does not cross-refer to any filing or licensing regime. The drafting explanation restates the clause with 及 in place of 以及, so the Article 8 text, not the explanation, is the locator for the exact words.

Identification, mode switching and the filing mechanism

Article 6 would require providers to have the capability to identify minor users, drawing on a list of methods: legal identity documents, the national network identity authentication public service, the distributed digital identity authentication public service, 非存储式一次性人脸核验 (non-stored one-time face verification) and behavioural characteristic identification. The face-verification method is one item on that list. The draft does not make it the sole or universal method, and it pairs the identification duty with a duty to protect minors' personal information. A second paragraph would have legal representatives handle phone-card applications for minors lacking full civil capacity, with telecom operators registering real identity information.

Article 9 is the operational hinge. For a user verified or identified as a minor, the provider would be barred from offering Article 4 services and would have to switch Article 5 services into minors mode in accordance with the provisions. It would also require an appeal channel, prompt acceptance and verification of appeals, and prompt lifting of the relevant measures for users verified as meeting the provisions' requirements. Article 7 adds, for minors mode specifically, a prohibition on presenting information that harms or may affect minors' physical and mental health, and a duty to intervene and contact a guardian or emergency contact when a minor user shows an extreme situation such as self-harm or suicide risk.

Articles 10 to 12 extend the scheme to hardware and distribution. Smart-terminal manufacturers would build minors mode linkage, one-touch switching, exit verification, anti-circumvention and controls over time, duration, function and content. App distribution platforms would assess age suitability, review apps seeking listing or updates under minors mode strictly, and restrict download and installation of apps that do not fit their age range. Both classes of actor would be bound by Articles 4 and 5 as if they were network service providers. Article 11 would set the functions minors mode has to have, with technical standards to follow. Article 12 would create a filing mechanism: providers of minors mode would file its settings and functions with provincial cyberspace departments, platforms with very large numbers of minor users and significant influence on minors would file through the provincial level to the national CAC together with a report on mode construction and usage, and filings would be verified annually. The telecom authority would add mobile-terminal minors mode to network access licence testing.

The penalty articles list six provisions, and Article 8 is not one of them

Article 18, as drafted, attaches penalties to violations of Articles 4, 5, 6(1), 7(3), 9(1) and 11(1). The enforcement bodies named are the cyberspace, press and publication, telecom, public security, culture and tourism, market regulation, and radio and television departments within their respective remits. The measures would be an order to correct, a warning and confiscation of illegal gains; where illegal gains are RMB 1 million or more, a fine of one to ten times those gains; where there are no illegal gains or they are under RMB 1 million, a fine of RMB 100,000 to 1 million; and a personal fine of RMB 10,000 to 100,000 on directly responsible managers and personnel. For refusal to correct or serious circumstances the draft adds business suspension, rectification closure, shutting the website or app, and licence or business-licence revocation.

Article 19 would cover the appeals duty in Article 9(2): an order to correct, and for refusal or serious cases a fine of RMB 50,000 to 500,000 on the entity and RMB 10,000 to 100,000 on responsible personnel. Article 20 would cover Article 10 and the filing duty in Article 12(2): correction, a warning, an optional fine of RMB 100,000 to 1 million, personal fines of RMB 10,000 to 100,000, and the same suspension and revocation powers in aggravated cases.

Article 8 appears in none of those lists. Article 21 routes other violations of the draft to existing law, naming the Minors Protection Law, the Cybersecurity Law, the Personal Information Protection Law, the Regulations on the Protection of Minors in Cyberspace and the Telecommunications Regulations, and preserves civil, public-security and criminal liability. That is the only route the draft gives for the algorithm clause. We therefore do not attach the Article 18 fine schedule to Article 8, and we do not say what a breach of Article 8 would cost.

What we did not verify

What we opened: the CAC notice at cac.gov.cn dated 18 September 2026, read end to end in the original Chinese, including the notice text with the comment channels and deadline, all 25 articles of the inline draft, and the drafting explanation in four parts. We verified the quotation character for character against the fetched bytes at 第八条. All translations in this piece are ours and are labelled as such.

What we did not open: the Minors Protection Law, the Cybersecurity Law, the Personal Information Protection Law, the Regulations on the Protection of Minors in Cyberspace, the Telecommunications Regulations, any existing national standard for minors mode, and the anthropomorphic AI interaction measures we link below. We describe the relationship between this draft and existing Chinese law only as the draft and its explanation describe it.

What we refuse to claim: we do not say the State Council has adopted or will adopt this regulation, or that it will take effect in 2026, because the notice is a consultation and Article 25 is blank. We do not say any obligation described here is in force. We do not say that every chatbot, recommendation system or education product reachable by minors in the PRC falls within Article 5(3), because that turns on the undefined phrase 可能影响未成年人认知 and on facts about each product. We do not say Article 8 requires proof of intent, or government approval of models, or that it carries the Article 18 fines. We do not say these are China's first protections of this kind or that they replace the July 2026 anthropomorphic AI measures, because the draft makes no such claim and we did not test it. We give no estimate of how many services would be affected, because the draft gives no number.

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.

Related briefings

• China's interim measures on anthropomorphic AI interaction services
• CAC draft Anti-Cyberviolence Law and its AI labelling duties
• CAC revised Internet Information Services Measures, AI section
• Asia-Pacific AI regulation tracker
• AI Regulation News hub