As of July 2026, 988 generative AI services in China had completed filing and 598 applications or features had completed registration, with 1,509 large models released; over the same period, first-phase enforcement had handled more than 14,000 non-compliant AI products. Put the two figures together and the message is clear: filing is not a formality.
Three routes, and the wrong one is easy to take
Algorithm filing, large-model filing and large-model registration are three different things: a self-developed or fine-tuned model goes through filing; an application that calls the API of an already-filed model goes through registration; and features involving recommendation or deep synthesis also need algorithm filing on top. The most common misunderstanding is “the model belongs to someone else, so filing is their business” — the model provider files, but the application provider still has to register, and one does not replace the other.
Content labelling is mandatory
The Measures for Labelling AI-Generated Synthetic Content and the supporting mandatory national standard GB 45438-2025 took effect on 1 September 2025, requiring both explicit labels (text, sound or graphic notices the user can see) and implicit labels (invisible marks in file metadata). Distribution platforms, app stores and users are all brought within the scope of obligation.
Enforcement is escalating
The “Qinglang · clean up AI application disorder” campaign launched in April 2026 puts “should have filed but did not” first among seven priority problems, and poor implementation of content labelling is on the list as well. In Shanghai a generative AI service has been formally placed on file for punishment after refusing to rectify, and 54 non-compliant AI apps have been removed from local app stores. Penalties escalate: warning and public notice, correction within a deadline, suspension of service, then administrative and criminal liability.
What it means for enterprises
- List your product lines first and decide, one by one, whether each is filing or registration; do not wait for an inspection to classify them.
- Check whether AI-generated content carries both explicit and implicit labels — the quickest item to fix, and a priority inspection point.
- Have the licensing documents for training data ready to produce; source compliance is a focus of checks.
- Publish the model name and filing (or registration) number prominently — this is an explicit requirement.
- For Hong Kong and overseas markets, local requirements stack on top; do not prepare only one set of documents.
Want to talk through what you just read?
We work end to end, from product design to launch. Tell us about your scenario and we can start there.
Shenzhen XZS Technology Co., Ltd. · Tel 0755-82494862 · 13760424861