China wrote a labelling rule for every generated video frame

The figure or the record

On 14 March 2025, China's Cyberspace Administration published, jointly with three other ministries, the 关于印发《人工智能生成合成内容标识办法》的通知 — the Notice on Issuing the Measures for Labelling AI-Generated and Synthetic Content, read here in the original Chinese. The Measures, signed 7 March 2025 by the Cyberspace Administration, the Ministry of Industry and Information Technology, the Ministry of Public Security and the National Radio and Television Administration, take effect from 1 September 2025, as their Article 14 states.

What the documents establish

The Measures define two label types: explicit labels (显式标识), meant to be visible or audible to a user, and implicit labels (隐式标识), embedded in file metadata and not easily perceived. Article 4 sets format-specific rules; for video specifically, a prominent prompt must appear on the opening frame and around the player during playback, and may also appear at the end or in the middle. Article 5 requires implicit labels to carry the content's attribute information and the service provider's name or code, and encourages a digital watermark on top of that. Article 7 requires app-distribution platforms to verify a developer's labelling materials before listing an app that offers generation services, and Article 10 bars anyone from maliciously stripping, altering or forging the labels the Measures require. A newer, sector-specific rule applies the same logic directly to micro-dramas: NRTA Order No. 16, effective 1 September 2026, states in its Article 34 that a micro-drama generated or produced using AI must carry a prompt label in a prominent position on every episode, in accordance with relevant national rules — an explicit tie-back to the CAC Measures' video requirement.

Scope and caveats

These are rules for labelling, not rules limiting what may be generated; nothing in either document opened for this record bans AI-made micro-dramas or requires disclosure of which specific tool produced one. The Measures place the labelling duty on the 'service provider' offering generation or hosting services, with a separate duty on distribution platforms to check for it; responsibility for a mislabelled or unlabelled finished episode circulating outside those channels is addressed by enforcement provisions this record has not traced to a specific case. The NRTA order's own effective date, 1 September 2026, falls only fifteen days before this record's retrieval date, so it has barely had time to generate an enforcement record.

What to watch

Editorially, whether app stores and short-video platforms are actually verifying labelling materials under Article 7, rather than relying on self-declaration, is the detail that will determine whether this rule changes visible practice or stays a paper requirement.

• Has any platform published a case of an app rejected, or a video removed, specifically for a missing AI label under these Measures?
• Does the NRTA's episode-level labelling duty produce a visibly different on-screen mark than the CAC's general video rule, in practice?
• How is an implicit, metadata-level label checked once a video is re-encoded or re-uploaded to a different platform?

A labelling rule is only as strong as the verification behind it; two ministries requiring a mark is not the same record as a platform enforcing one.