China Seeks Public Comments on Revised Internet Information Services Rules, Introduces AI Regulation Chapter

China released a draft update to its Internet Information Services Management Measures on July 3. The new version adds a full chapter titled "intelligent information services" that sets rules for AI providers.

The draft requires companies to publish basic technical principles, list sources of training data, and mark any generated or synthetic content. It also bans forcing users into AI services or using algorithms to shape online opinion. Feedback closes on August 2. The full text is available via the Cyberspace Administration of China (official announcement).

The draft brings AI under the same licensing and account rules that already cover other internet platforms.

This marks the first time Chinese regulators have grouped generative AI obligations into a general internet services law rather than creating a separate AI rule. Existing measures already require operators to hold an internet information service license. The new chapter extends those duties to any service that produces text, images, audio, or video through models.

Platforms must also maintain user account systems that allow deletion of accounts inactive for more than six months. They must build libraries of violent content features and offer users tools to block or stop reposts of such material.

Regulators say the changes address gaps that appeared after large language models became widely available in 2023. Earlier rules focused on recommendation algorithms and deepfakes. The new text treats generative output as a core service feature that needs consistent oversight.

Competing firms now face identical disclosure and labeling duties regardless of model size.

Large domestic developers such as Baidu, Alibaba, and Tencent already run licensed platforms and have published some safety reports. Smaller startups and foreign model providers must meet the same standards or partner with a licensed Chinese entity. The draft does not create different thresholds for model scale or compute use.

The text also reinforces the existing ban on using algorithms to amplify extreme positions or fabricate trends. Providers must keep records that allow authorities to trace how content reaches users.

The proposal keeps the focus on process transparency rather than outright content bans.

Draft language stops short of requiring pre-approval of every generated item. Instead it demands visible labels and accessible explanations of how models work. This approach mirrors earlier rules on recommendation systems that required platforms to offer users a choice to turn off personalized feeds.

International observers note that similar disclosure requirements appear in the European Union AI Act and in proposed U.S. state bills. The Chinese draft adds the extra constraint that training data sources must be listed publicly, a step not yet required in most Western proposals.

User rights receive clearer protections than in previous versions.

Account holders gain explicit rights to request deletion after six months of inactivity. Platforms must also supply blocking and reporting tools for violent or harassing material. The required feature libraries for violent content resemble tools already used by some social platforms to reduce resharing of harmful posts.

Critics question whether smaller operators can afford the compliance burden. Building labeled datasets and maintaining audit logs adds engineering overhead. Foreign companies without a local license must either apply for one or route services through a partner. The draft leaves open how enforcement will scale once the rules take effect.

Future compliance will depend on how strictly authorities interpret the new labeling and disclosure clauses.

The comment period ends August 2. After that, the Cyberspace Administration can issue a final version or further revisions. Companies already operating models inside China have begun mapping their current practices against the draft text. As noted by attorney Angela Zhang, a specialist in Chinese technology law at the University of Hong Kong, “These transparency requirements will force companies to document what previously remained internal, raising real compliance costs for smaller players” (Reuters).

Observers will watch three signals in the coming months. First, whether the final rule keeps the six-month account deletion clause or narrows it. Second, whether foreign providers receive any carve-outs or must fully localize. Third, whether any major platform publishes its first training-data summary ahead of enforcement.

Readers tracking AI policy should follow these updates because the rules affect any service that lets users generate text, images, or code within China or for Chinese users. The outcome will shape how openly companies describe their models and how visible synthetic content becomes on domestic platforms.