On 14 July 2025 the Cyberspace Administration of China published the twelfth batch of domestic deep synthesis service algorithm filings. The announcement is short, but it confirms once again an established logic: deep synthesis filing is not a one-off market-entry step but ongoing, batch-based administration.
Three filing actions: initial, change and cancellation
The announcement makes clear that providers of deep synthesis services with public-opinion attributes or social-mobilisation capacity must complete filing under the Provisions on the Administration of Algorithm Recommendations for Internet Information Services; where filed information changes, they must process a change filing promptly; and where a service ends, they must process cancellation.
These three actions form the full lifecycle of the filing duty. What is often overlooked in practice are the latter two — an algorithm version upgrade, an expansion of functional scope or a change of operating entity may trigger a change-filing duty, while taking a service offline without cancellation can leave a continuing exposure. See our note on generative AI filing and registration in local batches.
Technical supporters are in scope too
The announcement specifically notes that technical supporters of deep synthesis services must comply with the filing duty by reference. This matters greatly to the tools and capability providers in the middle of the industry chain: even without facing end users directly, a party that provides technical support for deep synthesis services may fall within the filing scope.
This explains why filing lists include both consumer-facing applications and providers of technical capabilities such as models, voice synthesis and image generation. See our note on the 13th batch of deep synthesis algorithm filings.
Dividing work between deep synthesis filing and generative AI filing
Companies often conflate deep synthesis filing with generative AI service filing. The two have different objects and trigger conditions: the former revolves around algorithms, the latter around services and model capability. See our note on generative AI filing under the 611 regime.
When a product has both generative and deep synthesis capability, it may need to satisfy both sets of requirements. This is also where on-device AI and face-recognition applications most easily miss a step in filing practice. See our notes on on-device AI filing and face recognition filing.
Reading the rhythm from the batches
After the twelfth batch came the thirteenth, published on 11 September 2025. Batches issued roughly two months apart show that the authority has settled into a stable rhythm of review and publication.
For companies, that rhythm is something to use: embed filing preparation into the product-iteration cycle and complete material updates before a version release, avoiding the compliance gap left by launching first and filing later.
Where companies can start
First, build an algorithm ledger recording each algorithm’s version, function, launch date and filing status.
Second, set change triggers, bringing major version upgrades and functional expansion into the internal change-filing process.
Third, clarify the technical supporter’s role, agreeing filing-cooperation duties with downstream partners in contracts.
The value of the filing system lies not in the list itself but in the way it turns algorithm governance into a trackable timeline. Whoever maintains that timeline can quickly account for itself when regulators come asking. See our note on China’s internet information service measures for intelligent services.
This article is based on public information on the Cyberspace Administration of China’s twelfth batch of domestic deep synthesis service algorithm filings and is for reference only; it does not constitute legal advice.