ChinaMeasures for Labeling AI-Generated Synthetic Content
In force since .
An AI transparency rule binding public and private bodies.
- Obligation class
- Content labelling
- Audit expectation
- none
- Setting the rule reaches
- General
What it requires
- Add a user-perceptible label to AI-generated text, image, audio, video, and virtual-scene content
- Embed a machine-readable identifier in file metadata carrying the generation attributes and your provider name or code
- Check incoming content for labels and add a risk prompt to unlabeled or suspected-generated content, if you distribute content others supply
- Never delete, alter, forge, or conceal a required label
These lines are LexLint's reading of the law's text.
In the frameworks
- NIST AI RMF
- MEASURE 2.8Risks associated with transparency and accountability – as identified in the MAP function – are examined and documented.
- NIST AI 600-1
- GAI-RISK-07Human-AI Configuration
- GAI-RISK-08Information Integrity
- OWASP Agentic Top 10
- ASI09Human-Agent Trust Exploitation
- MIT mitigations Draft
Our reading of where this law's duties sit in each framework. It never says that running a control meets the law. How we read a framework against the law
Who enforces it
Settledness
- As of
- Open questions
- Which generation services fall within 《互联网信息服务深度合成管理规定》第十七条第一款情形 ("the circumstances in Article 17, paragraph 1, of the Provisions on the Administration of Deep Synthesis Internet Information Services") so that Article 4 requires an explicit label?
- Does the Article 5 duty to add an implicit label to 文件元数据 ("file metadata") reach generated text that is delivered in an interface and never saved as a file?
What it makes you log
Logging duty
Articles 4 and 5 require an explicit, user-perceptible label and an implicit file-metadata label for generated or synthesized content; neither article states a duty to log, record, or later produce evidence that a labeling event occurred. A separate provision of the same Measures, Article 9, requires retaining logs of recipient information for no less than six months when a service provider supplies unlabeled content at a user's request, but that duty sits outside this instrument's own citation (Articles 4 and 5) and is not carried here.
- Kind
- None
- As of
What this law does
Internet information service providers must add an explicit, user-perceptible marker to AI-generated or AI-synthesized text, image, audio, video, and virtual-scene content, and must embed an implicit, machine-readable identifier in file metadata carrying generation attributes and the provider's name or code. Content-distribution platforms must check incoming content for labels and add a risk prompt to unlabeled or suspected-generated content.
No person or organization may maliciously delete, alter, forge, or conceal a required label. The accompanying mandatory national standard GB 45438-2025 sets the technical parameters for both label types and takes effect the same date; no official standards-body URL for GB 45438-2025 is available independent of a commercial reseller, so it is described here rather than authored as its own instrument.
When LexLint raises it
When your app profile says your app generates content with AI, generates text with AI, generates images with AI, generates video with AI, generates audio with AI or deploys a chatbot.
Back to the example · Lint your app
Page updated .