China's rules for visible and hidden labels on all AI content
The strictest AI content labelling regime of the major markets has been in force since September 2025.
Background
China has the most prescriptive AI content labelling regime of any major market. Since 1 September 2025, all AI-generated content distributed through covered services must carry both explicit and implicit labels.
Three key points
1. Two types of label are required. The Measures for Labelling AI Generated Content and national standard GB 45438-2025 require visible labels that audiences can see and embedded metadata labels that can be machine-read. Both types are required. Duties fall on service providers that generate the content and on distribution platforms that carry it.
2. Penalties for failures have been reported. It has been reported that the Cyberspace Administration penalised CapCut, Maoxiang and Dreamina AI in April 2026 for labelling failures, consistent with the enforcement pattern seen to date.
3. The standard is technical, not just editorial. The GB 45438-2025 national standard specifies technical requirements for the metadata label. Compliance requires implementation at the system level, not just a visible watermark added in post-production. Technology vendors rather than brands typically carry this obligation at the technical layer.
Why this matters
Content produced with AI tools and distributed in China, or on platforms serving Chinese audiences, must meet both label requirements. The rules sit alongside platform filing and safety review obligations. Checking compliance with Chinese AI labelling rules should be part of any due diligence on tools used for content destined for this market.
Sources: White & Case; Covington, Inside Privacy; SafeAI for Business.
