Under the Provisions on the Administration of Algorithm-Generated Recommendation Services and the Provisions on the Administration of Deep Synthesis Services, an algorithm filing is not a one-off act. A change in filed information, or termination of the service, requires a change or cancellation filing. By 2024 the algorithm filing system was operating routinely, and the continuing obligation has become much more visible. Product iteration and filing maintenance need to sit on the same timetable.
What triggers a change
In practice, changes often arise from a new entity name or corporate detail, an adjustment to the algorithm’s name or main function, a change in the product or launch channel, or a change in service recipients and scenarios. Companies often treat product iteration as purely technical work and overlook a substantive change in the algorithm’s mechanism. The test is not how many lines of code changed, but whether the externally presented service and recommendation logic changed. A redesign of the interface may not matter; a shift in recommendation objectives may.
Deadlines and materials
Change and cancellation must be completed within the prescribed period. Materials usually include a change note, an explanation of the algorithm mechanism, a safety-assessment conclusion and supporting documents. The key is comparability between versions: showing reviewers what changed and what it affects is more effective than stacking technical documents. Where one entity holds several algorithm filings, each should be checked separately so that an update to the main product does not leave ancillary functions behind.
Consequences of not filing
A mismatch between filed information and actual service surfaces in inspections, complaint handling and takedowns. The deeper effect is that filing status is connected to the launch of generative AI services, app-store listing and the conditions for platform cooperation. Treating a change as something that can be postponed tends to produce a cluster of costs during expansion. A small compliance gap can also be magnified in financing due diligence or partnership negotiations.
Interaction with generative AI filing
Generative AI services with public-opinion attributes or social-mobilisation capacity also involve the interaction between generative AI service filing and algorithm filing. The two focus on different things: the former on service and safety assessment, the latter on algorithm mechanism and entity information. Where a product has both recommendation and generation functions, the two sets of materials should sit in one internal register to avoid version drift. External descriptions should also be consistent, so that the same function is not described differently in different materials.
Building an internal register
At a minimum, record each filing’s number, version, corresponding product, launch channel, responsible person and date of last change. The register helps identify the scope of impact when a change occurs and prevents information from disappearing when staff leave. For companies with multiple product lines, the register should be tied to the release process: confirm filing status before a version goes live, not after. The register should also keep historical versions so that a regulator can be shown the trail of changes, rather than only the current state.
For filing changes, cancellations and material review, see our algorithm filing service and data compliance service; for synthetic-content labelling, see AIGC labelling service.
This article is based on the Provisions on the Administration of Algorithm-Generated Recommendation Services, the Provisions on the Administration of Deep Synthesis Services and public information about the algorithm filing system in 2024. It is for reference only and does not constitute legal advice.