(Yicai) Sept. 2 -- New Chinese regulations requiring all artificial intelligence-generated content to be labeled came into force yesterday following a six-month grace period.

All AI-generated audio, images, text, video, virtual scenes, and related outputs must carry both visible labels and embedded metadata tags, according to the regulations issued in March. The new rules apply to AIGC platforms, service providers, and users on content sharing and social media sites.

The authorities pointed to the rapid growth of generative AI and deep synthesis, which can create realistic text, voice and imagery, as the reason for the move, citing the risks of misinformation, fraud, and impersonation.

AI platforms and service providers must now label AIGC so that people can tell when something was created by AI and where it came from, Li Hongda, chief engineer at the Shanghai Information Security Testing Evaluation and Certification Center, said at an event yesterday.

Content sharing platforms, such as short-video site Bilibili and social e-commerce app Xiaohongshu (Rednote), must check the metadata of the AIGC that users publish and add interface tools that prompt or require users to label AI-generated material.

Bilibili has informed users about the new regulations and its AI-labeling feature a number of time through announcements, site messages, pop-ups and explanatory videos, said Luan Shuai, head of content production at the Shanghai-based company.

Zhou Xiaorong, chief editor at AI model developer MiniMax, demonstrated at the same event yesterday how text, images, voice, and video generated using its products -- MiniMax, Hailuo, MiniMax Audio, and Xingye -- are clearly labeled with AI-generated watermarks.

The Shanghai branch of China’s cyberspace watchdog also established an alliance of over 30 AI companies to jointly carry out mutual recognition and verification of embedded AIGC labeling.

Editor: Futura Costaglione