China's Mandatory AI Content Labeling Era Begins: Rules in Force September 1, 2025

The Measures for Labeling AI-Generated and Synthetic Content — jointly issued by the Cyberspace Administration of China, MIIT, the Ministry of Public Security and the NRTA — took effect on September 1, 2025: all AI-generated text, images, audio and video must carry explicit labels (user-visible) and implicit labels (metadata watermarks).

It is the world's first fully implemented mandatory AI-content labeling regime: generation platforms, distribution platforms and app stores each carry obligations, and removing labels or offering label-stripping tools is a violation. WeChat, Douyin and others rolled out AI-content declaration features in step.

With deepfakes and AI rumors plaguing the globe, China's labeling regime — alongside the EU AI Act's transparency rules and U.S. state deepfake laws — forms one of three paradigm samples for global AI content governance.

Explicit Plus Implicit: The Dual-Label Logic#

The Measures require explicit labels (user-visible badges/disclosures) and implicit labels (metadata watermarks) in parallel, targeting the same pain point: explicit labels are easy to crop out, while implicit watermarks are hard to notice and easy for platform moderation to trace. Generation platforms, distribution platforms and app stores each bear obligations, and removing labels or offering stripping tools is a violation — a chained-liability design far harder to evade than single-point labeling.

The Divergence of Three Paradigms#

Global AI content governance is splitting along three routes: China on 'content labeling' (mandatory, whole-chain), the EU on 'risk tiering' (AI Act transparency), the U.S. on 'state deepfake laws' (fragmented, litigation-driven). This tracks the post-Paris US-EU-China regulatory divergence (see our coverage) — China's labeling regime is the world's first fully implemented mandatory sample, and its enforcement will be a key reference for other legislatures.

Our Take#

Content labeling turns 'AI-generated' from a gray zone into a legal disclosure duty — essentially rebuilding trust infrastructure for the AI era. For exporters and content platforms, compliance is no longer optional: adapting labeling rules by jurisdiction and building in watermarking is becoming product infrastructure rather than an add-on — and the fragmentation of the regulatory map is itself a new cost of doing business.