On 18 September 2026 the Cyberspace Administration of China (CAC) published a 25-article draft State Council rule on minors' healthy and safe use of the internet. Comments closed on 17 October. The effective date in Article 25 is still blank, and the final text may change.
The strongest case for the draft
The case for acting is serious. Children are not small adults. Chinese commentators quoted by China Daily argue that under-16s "are generally not mature enough and lack sufficient life experience" to host livestreams. Liu Ying of the Chinese Academy of Cyberspace Studies said that minors' risks vary with age. A rule that varies obligations by age is more refined than a flat ban. Products that simulate a parent, partner or relative and are tuned to keep a lonely teenager talking carry a plausible risk of emotional dependence. Regulators acting before harm is documented is a defensible position.
What the draft actually does
The CAC consultation text sets out the following:
- Article 4 bars providers from offering minors virtual intimate-relationship services, "virtual relatives" and "virtual partners" among them. It also bars livestream publishing services to under-16s and stranger social networking for minors. The one exception is for those over 16 who mainly support themselves through their own earnings.
- Article 5 requires under-16s to reach other livestream services, online games, social networking and certain AI services through a "minor mode".
- Article 6 requires providers to identify minor users, using methods such as legal ID documents, national online identity services and one-time facial verification.
- Article 8 requires algorithm review and management systems. It forbids algorithm models that induce emotional dependence, addiction or excessive spending in minors.
- Articles 10 to 12 make device makers offer minor-mode linkage and one-touch switching. Providers must file minor-mode details with provincial regulators, and large platforms file nationally.
- Article 18 sets fines of 100,000 to 1 million yuan for most violations, or 1 to 10 times illegal gains where those gains reach 1 million yuan. Responsible individuals face 10,000 to 100,000 yuan, and serious cases can end in suspension or licence revocation.
The draft sits on top of the Regulations on Online Protection of Minors, a 60-article rule in force since 1 January 2024. It also follows the Interim Measures for the Administration of Anthropomorphic Interactive Services, which took effect on 15 July 2026 and, according to Jiemian, already bar virtual relatives and partners for minors. The virtual-companion ban is therefore less a new idea than a move of an existing prohibition into the general rulebook.
Where the draft is well built
Two features deserve credit. First, Article 9 requires an appeal channel and requires providers to lift restrictions for users who prove they qualify. Most age-assurance regimes skip this step, and it recognises that classification errors are inevitable. Second, the under-16 livestream rule targets hosting, not watching. A minor can still view, within a minor mode. That is a narrower measure than the blunt access bans being debated elsewhere.
Where proportionality breaks down
Identification is the hidden cost. A provider cannot apply a minor mode to under-16s without knowing who is under 16. The draft does not say that every user must verify, but that is the practical consequence of a duty to identify minors. This is an inference from the text, not an explicit provision. Building a reliable age signal means collecting ID or facial data from adults as well. That creates a large store of sensitive data, and it ends pseudonymous participation for everyone. Child-safety rules that are paid for with adult anonymity are a trade the draft makes without naming it.
The algorithm clause is vague. An algorithm that does not "induce" dependence, addiction or "excessive" spending is a standard with no measurable threshold. Engagement tuning is common in games, video feeds and chat products alike. Without a defined test, providers will over-comply. They will strip personalisation and features for minors, and perhaps for anyone they cannot confidently classify as adult. Penalties of up to a million yuan, plus personal liability for managers, push in the same direction.
Bright-line bans substitute for evidence. The draft cites no evidence on what share of harms come from virtual companions or from stranger chat. A proportionate regime would pair the ban with a requirement to report incident data, so regulators can tell whether the rule works. The filing duty in Article 12 collects information for the authorities. It is not a public evidence base.
Innovation cost. Companion and AI-character products are a fast-growing area where Chinese firms compete globally. A ban on minors is defensible. But a categorical rule on "virtual intimate relationships" with unclear edges could catch tutoring avatars, language-practice partners or mental-health tools. Clear definitions and safe harbours for these uses would cost regulators little.
What a better final text would do
The draft is more careful than the most restrictive models. It still leaves four fixes for the final version:
- Say plainly that age assurance should use the least data needed, and that facial or ID data must be deleted after the check.
- Replace "induce" and "excessive" with testable criteria, or at least with regulator guidance on safe design.
- Define "virtual intimate relationship" so that educational and therapeutic uses are clearly outside it.
- Require published, aggregate outcome data, so the rule can be judged on results.
The consultation has closed, and the CAC's revisions will show whether protecting children and preserving adult privacy can be held together, or whether the first is bought with the second.