Chinese Government Policies Regulating AI

China's first specific administrative regulation on generative AI services, jointly released by the Cyberspace Administration of China and six other government departments. The measures establish requirements for generative AI service providers, including content monitoring, data security, and algorithmic transparency. They also prohibit the generation of content that endangers national security or promotes discrimination.

These provisions regulate the use of deep synthesis technology (such as deepfakes) in internet information services in China. They require deep synthesis service providers to verify users' real identities, conduct security assessments, and implement content review mechanisms. The regulations prohibit the use of deep synthesis to create, copy, publish, or disseminate illegal information or content that may endanger national security or damage the public interest.

These provisions regulate the use of algorithmic recommendation technology in internet information services in China. They require algorithm providers to promote "mainstream values" and prohibit using algorithms to engage in activities that endanger national security or disrupt economic and social order. The regulations also mandate that algorithm providers inform users about the basic principles and purpose of their algorithms and provide users with options to turn off algorithmic recommendations or select preference parameters.

Released by China's National Technical Committee 260 on Cybersecurity, this framework provides guidance for AI safety governance. It emphasizes a people-centered approach and the principle of developing AI for good. The framework outlines principles for AI safety governance, including balancing development and security, ensuring fairness and transparency, and promoting international cooperation. It also establishes requirements for AI safety management, including risk assessment, monitoring, and emergency response.

These upcoming rules will make it mandatory for AI-generated content to be labeled, both explicitly and implicitly. Explicit labels must be easily perceived by users and added to text, audio, images, videos, and virtual scenes, while implicit labels are embedded within the file's metadata. The regulations aim to help users distinguish between AI-generated and human-created content, thereby reducing the risks of misinformation and deepfakes.

The State Administration for Market Regulation and the Standardization Administration of China have released three national standards for enhancing the security and governance of generative AI:

1. Cybersecurity Technology—Generative Artificial Intelligence Data Annotation Security Specification, which outlines security requirements for data labeling processes
2. Cybersecurity Technology—Security Specification for Generative Artificial Intelligence Pre-training and Fine-tuning Data, which establishes requirements for ensuring the security of datasets
3. Cybersecurity Technology—Basic Security Requirements for Generative Artificial Intelligence Service, which sets security requirements for generative AI services, including user data security assessments and protection measures