a mandatory mark on content generated with artificial intelligence

For the first time, a project contains specific rules for the sector, both companies and content creators. Videos must contain an “explicit” tag. Specialists anticipate difficulties in the application.

Beijing () – Beijing has proposed new regulations that establish the obligation to clearly identify all content generated by artificial intelligence, in an attempt to stop the increase in fraud related to the use of this new technology. The draft containing the guidelines was published on September 14 and was open to public comments for a month. Requires all AI-generated images, videos, and audio to be clearly labeled with a watermark (or electronic watermark) and embedded metadata. It is the first time that the Chinese Cyberspace Administration proposes specific rules for the definition and classification of content generated by artificial intelligence.

According to the guidelines, generated videos must contain an “explicit” label at the beginning of the clip and visible at all times in a corner of the screen. There is also a strong recommendation to flash the tag at certain “appropriate” moments in the same video. Metadata about the file’s source and copyright information – indicated as an “implicit” tag – must also be recorded at the time of creation.

The rules would apply not only to artificial intelligence companies, but also to individual content creators, online platforms, app stores and any other content distributor. According to the same guidelines, content distribution platforms are required to label files suspected of having been generated by artificial intelligence, in case the metadata is missing. Finally, app stores must ensure that content providers correctly label AI-generated content.

The proposals are the latest effort by Chinese authorities to combat the rise in fraud cases related to the new technology. According to Chinese startup RealAI, more than 185 million yuan ($26 million) was stolen in China using AI in the first 5 months of 2024, compared to just 16.7 million yuan in all of 2023. The Technology Deepfake, text-to-speech, and AI chatbots are defined as “typical high-risk application scenarios,” highlighting their potential use to defraud users.

However, analysts and experts point out that the new directives may not be so easy to apply in practice. Among the critics is Ma Ce, a lawyer at Zhejiang-based Internet law firm Kinding, who says the guidelines lack a clear definition of what constitutes “AI-generated content that requires watermarking.” The obligation imposed on creators and online platforms to correctly add metadata to generated files “will pose technical challenges,” the expert added. The directives also do not specify sanctions for content creators and distributors who violate the rules.

China is not the only country that is trying to ensure that AI-generated content is clearly labeled. The European Union, the United States, Singapore and Canada are taking steps to regulate this sector. However, several questions are raised about how the rules requiring watermarks or watermarks on online content will be implemented.