Algorithmic recommendation and filing

Endless Feeds and Minors: What Do the Rules Require?

What you may need to do

1. Review and assess models and data periodically, screening for addiction-inducing design
2. Provide an age-appropriate mode for minors and have it on by default
3. Exclude content that may invite imitation of dangerous behaviour or bad habits from the recommendation pool for minors
4. Build monitoring and alerting for excessive-spending risk

Clause basis

One-line conclusion

Two red lines: first, algorithm models must not be set up to induce users to become addicted or to overspend; second, when providing services to minors, the provider must both offer an age-appropriate mode and refrain from pushing content that may cause minors to imitate unsafe behaviour or induce bad habits, and must not use algorithms to induce minors to become addicted to the internet.

Basis

The underlying obligation in Article 8 is periodic review: an algorithmic recommendation service provider shall periodically review, assess and verify the mechanisms of its algorithms, as well as its models, data and application results, and may not set up algorithm models that induce users to become addicted or to overspend, or that otherwise violate laws and regulations or ethical morality. Note that the first half is a process obligation and the second half is a prohibition on outcomes; both must be met at the same time.

Article 18 is directed at minors and has two paragraphs: the first requires that, where services are provided to minors, the provider performs its duty of online protection for minors in accordance with law and, by developing a mode suited to minors' use and providing services suited to minors' characteristics, makes it easier for minors to obtain information beneficial to their physical and mental health; the second is a prohibition, covering information that may cause minors to imitate unsafe behaviour or conduct that violates social morality, or induce unhealthy habits, and that may affect their physical and mental health, and the use of an algorithmic recommendation service to induce minors to become addicted to the internet.

Article 32 brings these two articles into the responsibility-linkage clauses: a breach is handled under the provisions of the relevant laws, administrative regulations and departmental rules.

What you need to do

• Build a periodic algorithm review mechanism, with "does it induce addiction, does it induce overspending" as fixed review items.
• Provide an age-appropriate mode for minors, and make the default state and the way out explicit.
• Exclude, at the recommendation-pool level, the content categories that may invite imitation of dangerous behaviour or bad habits.
• Monitor signals such as continuous session length, night-time use and repeated gifting, and set alerts.
• Keep review conclusions and rectification records on file, so they can be explained during supervision and inspection.

Common pitfalls

Making the minors mode a secondary entrance that users have to turn on themselves. Driving high-frequency use with mechanics such as check-ins, share-the-pot bonuses and limited-time leaderboards while never assessing addiction risk at review. Changing only the recommendation pool in minors mode, while the leaderboards, comment section and direct-message entrance remain the same as the adult version. Writing periodic review into the policy with no review record you can actually produce.

Common follow-up questions

The clause does not set the age boundaries of a minors mode or list its specific functions; implementation details go by the local cyberspace administration office's position.

Common follow-up questions

Must a minors mode be on by default?

Article 18 requires making it easier for minors to obtain beneficial information, including by developing a mode suited to their use; it does not directly regulate the default state, so discuss implementation with the local cyberspace administration.

How is inducing addiction determined?

Article 8 prohibits setting up algorithm models that induce addiction or excessive spending at the model level; the specific finding falls within regulatory discretion and this knowledge base does not speculate about it.

This entry belongs to the China AI Compliance & Filing cluster (42 entries in total): View all · Answers home