On 12 June 2024 the Cyberspace Administration of China published the sixth batch of domestic deep synthesis service algorithm filing information, searchable through the internet information service algorithm filing system. A batch announcement carries limited information in itself, but the rhythm it reveals matters: filing is moving from individual submissions to a predictable, routine process.
How to read a batch announcement
The announcement provides a list of filing entities and algorithm names; the mechanism descriptions are not published. What a company can observe is which application scenarios continue to be filed, which types of algorithm recur and how filing entities are distributed across industries. This structural information is more useful than any single entry. Comparing several batches in sequence shows where supervisory attention is moving.
Entity and algorithm granularity
One entity may hold several algorithm filings, and granularity depends on product form and service method. Whether to package several functions into one filing or split them requires a judgment about the actual service boundary. Too coarse, and a change becomes hard to map; too fine, and maintenance cost rises. For a platform offering face-swapping, speech synthesis and text generation together, splitting by functional module usually makes later changes easier.
Change and cancellation as continuing duties
After filing, changes to entity information, algorithm function or service scenario may trigger a change filing; cessation of service requires cancellation. Batch announcements make registered status more transparent and inconsistencies easier to detect. A company should assign an internal owner to maintain the filing register and link product versions to filed information, so that updating materials is not an afterthought once a version has shipped.
Labelling and filing
Deep synthesis services also involve content-labelling requirements. Filing and labelling are different compliance actions. Filing addresses the registration of the entity and algorithm; labelling addresses the identifiability of content. Both point to the same risk: synthetic content mistaken for real. Companies should embed labelling rules in the generation process rather than adding labels manually at publication. A platform offering both templates and tools must also distinguish its own generated content from user-generated content and state that division in the user agreement. Missing labels tend to be amplified in distribution, and later deletion does not remove the impact. For a new service, filing rhythm also affects the launch window; building batch-announcement timing into product scheduling reduces uncertainty.
Division of labour with generative AI filing
Deep synthesis algorithm filing and generative AI service filing focus on different things. The former addresses algorithm mechanism and entity information; the latter addresses service form and safety assessment. Where a product has both generation and synthesis capability, the two sets of materials should share the same factual basis, so that one function is not described inconsistently in different filings. External descriptions of functionality should also follow a single line.
For filing and material consistency review, see our algorithm filing service and AIGC labelling service; for data processing arrangements, see data compliance.
This article is based on the sixth batch of domestic deep synthesis service algorithm filing information published by the Cyberspace Administration of China on 12 June 2024, and on public information. It is for reference only and does not constitute legal advice.