China AI Labelling.

Use. explicit + implicit labelling of AI-generated content (in force 2025-09-01).

Register. 09 · Compliance controls — one of the 17 registers of the PLANiSPHERE corpus library.

China AI Labelling is enacted (or near-enacted) law in a specific jurisdiction, with a specific obligation and a specific date. Planisphere maps the obligation to an evidence record supporting the bias audit, transparency disclosure, or high-risk notice — and pins it to the section and effective date so a third party can verify it.

Answers: What does China AI Labelling require, by when, and what does Planisphere deliver for it?

Planisphere measures your tool deployment against China AI Labelling and seals the result into a signed, Merkle-rooted evidence record. The grade recomputes on your own hardware; the model state never crosses the boundary.

The Measures for Labelling AI-Generated Synthetic Content, issued in March 2025 by the Cyberspace Administration of China with MIIT, the Ministry of Public Security and the NRTA, took effect 2025-09-01 alongside a mandatory national standard on labelling methods. They bind four roles, not one. A generation service provider adds an explicit label a person can perceive to generated text, images, audio, video and virtual scenes, and that label must survive downloading, reproduction and export (Art. 4); it also writes an implicit label into file metadata carrying the content's attribute information, the provider's name or code, and a content reference number (Art. 5). A propagation platform verifies that metadata, adds its own platform name or code and content reference number, and shows a conspicuous notice around the published content (Art. 6). App distribution platforms check whether an application provides generative AI services and review its labelling materials (Art. 7). Publishing users must proactively declare generated content and use the platform's labelling function (Art. 10). Unlabelled supply is permitted only after the service agreement fixes the user's own labelling obligations, and the relevant logs must be retained for at least six months (Art. 9).

What the evidence record proves.

• That an explicit label was applied, with its form, placement and modality
• That an implicit metadata label was embedded, committing the provider code and content reference number
• That a distributing platform added its own identifiers and the notice it showed
• That a publishing user's declaration was received
• The unlabelled-supply agreement, a commitment to the recipient's identity, and the retention horizon

What this does not prove. Planisphere does not label, watermark or detect content, and does not host or distribute it. The record is evidence that the operator performed the labelling, notice and declaration acts — not an algorithm filing, not a security assessment, and not a determination that a label satisfies the national standard.

Cite-anchor: Measures for Labelling AI-Generated Synthetic Content (CAC, MIIT, MPS, NRTA) · Arts. 4-7, 9-10; with the mandatory national standard on labelling methods

Any internal link in the Planisphere site that names "China AI Labelling" canonicalises here.