Measures on generative AI and content labelling
Interim Measures for Generative AI Services (2023) and Measures for Labelling AI-Generated Content (2025) · Official source (CAC (in Chinese))
China has no general AI law, but was among the first countries to regulate it through targeted measures: recommendation algorithms (2022), deep synthesis (2023), generative AI (2023) and labelling of generated content (2025). Since 2026 the Cybersecurity Law also contains AI provisions.
⚠️ A general AI law is in the 2026 legislative plan only as a preliminary study item. Rules change often (e.g. new 2026 rules on "anthropomorphic" and companion AI): check the updated texts on the CAC website.
Who it applies to
- Providers of generative AI services to the public in China, including foreign providers
- Providers of recommendation algorithms and deep synthesis services
- Platforms distributing AI-generated content
- Excluded: research and development not offered to the public
Key obligations
Algorithm filing
Algorithms capable of influencing public opinion must be filed with the CAC; generative services are subject to a security assessment.
Content control
Generated content must not violate Chinese law and must respect "core socialist values".
Training data
Lawful sources, respect for intellectual property and personal data, measures for quality and accuracy.
Labelling
Visible labels and implicit metadata labels for AI-generated text, images, audio, video and virtual scenes (standard GB 45438-2025).
User protection
Clear terms of use, complaint mechanisms and anti-addiction measures, especially for minors.
Penalties
CSL, DSL and PIPL penalties
Violations are punished under the general laws (e.g. up to RMB 10M under the Cybersecurity Law) and may lead to service suspension.
Competent authorities
Cyberspace Administration of China (CAC), with the Ministry of Industry (MIIT) and the Ministry of Public Security (MPS).