On 12 March 2026 the Cyberspace Administration of China (CAC) published the sixteenth batch of domestic deep-synthesis service algorithm filings, covering 779 algorithms. Public information shows that the Doubao enterprise voice-podcast model algorithm, the Qwen image-recognition-generation algorithm and the Turing Zhixin text-generation algorithm were among them.
What the rhythm tells us
Judged by batch numbers, deep-synthesis algorithm filing has settled into a steady rhythm. The fifteenth batch was published in January 2026, the sixteenth in March. Both the interval and the volume indicate this is a routine registration mechanism, not a one-off clean-up.
More meaningful is the widening scope of algorithms brought into filing. From early face-swap and voice-synthesis capabilities to today’s image-recognition-generation, text generation and voice-podcast models, deep-synthesis filing is converging on “generative capability itself”.
Who should pay attention
Those directly providing deep-synthesis services. Services able to generate or edit faces, voices, text, images or video typically fall within the Deep Synthesis Provisions. Filing is a necessary step before launch — see our algorithm filing service.
Those embedding deep-synthesis capabilities in products — marketing tools, customer-service systems or content platforms integrating face-swap or voice-cloning features. Here a company must determine whether it is a service provider or a technical supporter; the duties differ.
Those providing generative AI services. Deep-synthesis filing and generative AI service filing are two separate procedures that are easily confused — see our note on filing vs registration for large models.
Self-inspection points
First, verify the product against the filed information. Where product features exceed the capability described at filing, assess whether a variation or new filing is needed.
Second, implement the labelling duty. Deep-synthesis content must be labelled as required, and failures are already within enforcement sight — see our note on the CapCut enforcement case.
Third, distinguish on-device from cloud. On-device generative capabilities have their own filing approach — see our note on on-device generative AI filing.
Fourth, keep cross-batch consistency. Algorithms filed across several batches should keep their naming, capability descriptions and use scenarios internally consistent, avoiding explanation costs during review.
Filing itself is not complex; keeping filed information in step with a fast-iterating product is.
This article is based on the CAC’s sixteenth batch of deep-synthesis service algorithm filings published on 12 March 2026. It is for general reference only and does not constitute legal advice.