China issues mandatory AI-generated content labelling rules

The Cyberspace Administration of China, alongside the Ministry of Industry and Information Technology and other regulators, finalised the Measures for Labeling of AI-Generated Synthetic Content, together with a mandatory accompanying national technical standard, GB 45438-2025, specifying how the labelling must work in practice. The rules take effect on 1 September 2025.

The measures require two layers of labelling on AI-generated text, images, audio and video: an explicit label visible to users, typically wording such as “AI-generated” placed at the start, end or another prominent position in the content, and an implicit label embedded as machine-readable metadata — including the identity of the service provider and a content reference number, with watermarking required where technically feasible. The obligation falls on both the internet information services that generate the content and the platforms that distribute it, which must run automated checks to detect the metadata and verify labelling compliance, with the CAC, MIIT and the National Radio and Television Administration empowered to audit and enforce.

The rules placed China among the first major jurisdictions to mandate machine-readable AI content provenance at the infrastructure level, rather than relying on voluntary industry watermarking standards such as C2PA. They followed several years of narrower Chinese rules targeting deepfakes and synthetic media specifically, extending the labelling requirement to the broader category of AI-generated content across text, image, audio and video regardless of whether it was used deceptively.