The news outlet China Daily reported on Friday that China will require all AI-generated content to be labelled from September 1st, 2025.

China Daily reports:

Chinese authorities issued guidelines on Friday requiring labels on all artificial intelligence-generated content circulated online, aiming to combat the misuse of AI and the spread of false information.

The regulations, jointly issued by the Cyberspace Administration of China, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the National Radio and Television Administration, will take effect on Sept 1.

A spokesperson for the Cyberspace Administration said the move aims to “put an end to the misuse of AI generative technologies and the spread of false information.”

According to China Daily, “[t]he guidelines stipulate that content generated or synthesized using AI technologies, including texts, images, audios, videos and virtual scenes, must be labeled both visibly and invisibly” (emphasis added by IPTC). This potentially means that IPTC or another form of embedded metadata must be used, in addition to a visible watermark.

“Content identification numbers”

The article goes on to state that “[t]he guideline requires that implicit labels be added to the metadata of generated content files. These labels should include details about the content’s attributes, the service provider’s name or code, and content identification numbers.”

It is not clear from this article which particular identifiers should be used. There is currently no globally-recognised mechanism to identify individual pieces of content by identification numbers, although IPTC Photo Metadata does allow for image identifiers to be included via the Digital Image GUID property and the Video Metadata Hub Video Identifier field, which is based on Dublin Core’s generic dc:identifier property.

IPTC Photo Metadata’s Digital Source Type property is the global standard for identifying AI-generated images and video files, being used by Meta, Apple, Pinterest, Google and others, and also being adopted by the C2PA specification for digitally-signed metadata embedded in media files.

According to the article, “Service providers that disseminate content online must verify that the metadata of the content files contain implicit AIGC labels, and that users have declared the content as AI-generated or synthesized. Prominent labels should also be added around the content to inform users.”

Spain’s equivalent legislation on labelling AI-generated content

This follows on from Spain’s legislation requiring labelling of AI-generated content, announced last week.

The Spanish proposal has been approved by the upper house of parliament but must still be approved by the lower house. The legislation will be enforced by the newly-created Spanish AI supervisory agency AESIA.

If companies do not comply with the proposed Spanish legislation, they could incur fines of up to 35 million euros ($38.2 million) or 7% of their global annual turnover.