China penalized CapCut and Dreamina over AI labels: what to check

In April 2026 China's internet regulator, the Cyberspace Administration of China (CAC), penalized the apps CapCut, Maoxiang (Cat Box) and Dreamina AI for failing to properly label AI-generated or synthesized content. If you pick an AI video tool that serves viewers in China, the lesson is that the regulator went after the tool providers, so ask a vendor how it labels before you assume it does.

The facts below come from TechNode's report of the action and Inside Privacy's summary of the labeling rules, both read on 2026-10-02. I have not read the CAC notice itself, so treat the details as the press accounts give them.

What exactly did the CAC do?

TechNode, dated 29 April 2026, reports that the CAC instructed local regulators to impose regulatory interviews, rectification orders, formal warnings and stricter accountability for responsible personnel on the three apps. It names three legal bases: the Cybersecurity Law, the Interim Measures for the Administration of Generative Artificial Intelligence Services, and the Provisions on the Identification of AI-Generated Synthetic Content.

A CAC official is quoted saying all online platforms must strictly adhere to the legal requirements and that there is no room for compromise or circumvention. The report lists no fine amounts, so do not assume a number.

What do the labeling rules ask of a video tool?

Inside Privacy describes two label types that took effect on 1 September 2025. Explicit labels are visible indicators in text, audio or graphics that tell a user the content is AI-generated. Implicit labels are metadata embedded in the file that carry the service provider's name and a content ID.

The obligated parties are internet information service providers and online content distribution services, such as social platforms, that create or distribute AI-generated content. That is why the tools themselves were penalized: they are the party producing the file.

Does this reach a team that only calls an API?

Not directly in the sources I read. The reports describe duties on service providers and distribution platforms, and they say nothing about a customer who renders a clip through a third-party API and posts it. What you can say is that a platform that distributes your clip is under its own labeling duty, so an unlabeled upload is more likely to be flagged or relabeled there.

If you publish into China, get advice from counsel on your own role. Do not read this post as a statement that API customers are or are not covered.

What can you do with Sume today?

Sume's docs describe what each tool returns and say nothing about a watermark or embedded provenance record on outputs, so treat the output as unmarked until you have checked the delivered file yourself. That means the implicit metadata label is yours to arrange outside Sume, and the Sume docs give no promise that a trim or a Timeline render keeps metadata a source file carried.

The explicit, visible label is the part you can do inside a Sume job. Video captions takes cues with text, start and end in seconds and burns exactly that copy, which works for a short on-screen AI-generated notice. Timeline compose can place a still, such as a prepared label graphic, on screen with a video for one shot. Both return a new MP4 on a job you can read through Jobs and results.

Sume does not translate the legal wording for you, does not generate the Chinese label text a regulator expects, and does not write the provider name or content ID into file metadata. Those steps stay with you.

A short vendor checklist

Use this as a working list and adjust it to your own pipeline and counsel's advice.

• Ask whether the tool burns a visible AI label and whether you can turn it off.
• Ask whether the file carries a provider name and content ID in metadata, and check the delivered file, not the preview.
• Re-check after every re-encode, because trimming, resizing or compositing can drop metadata.
• Keep the job id and delivered file URL for each clip so you can show how it was made.
• Re-read the CAC notice yourself before relying on this summary.