Chinese AI governance framework detailed

A European company using AI agents in China now understands the regulatory framework Beijing plans to apply, though it is unclear whether these agents will be classified under a stricter "sensitive areas and key sectors" regime or a less restrictive "low-risk areas" regime. The framework involves requirements such as registration, testing, and product recalls for the former, and self-assessment for the latter.

Notice outlines 18 measures for AI development

A notice titled making commerce prosper through technology (科技兴商), published on October 19, includes 18 measures aimed at promoting AI development in areas like flexible manufacturing, home services, logistics, and e-commerce. The document was jointly issued by nine government bodies, including the Ministry of Commerce and the China Cyberspace Administration.

New business models and security regulations

The tenth measure of the notice calls for the development of new business models centered around commercial AI agents, such as intelligent purchase guides and logistics systems. The fifteenth measure focuses on improving the governance system for security and regulation, including data collection rules and a monitoring system for data risks.

Two regulatory regimes for AI agents

The eleventh measure divides AI agents into two categories: "sensitive areas and key sectors," which require registration and testing, and "low-risk areas" like daily life or office work, where self-assessment is sufficient. The fourteenth measure suggests a reliability rating system by professional organizations to evaluate AI functions such as "consumption encouragement" and "false advertising."

China's AI governance builds on existing models

The October notice applies an existing governance model to commercial AI, including a negative list of data uses and sandbox testing for new technologies. It is part of broader efforts to "optimize the development environment of commercial technologies," including support for startups and data infrastructure.

EU AI Act contrasts with Chinese approach

The European Union's AI Act categorizes AI systems into four risk levels, from "unacceptable" to "minimal," with high-risk systems subject to stricter rules. The regulation, which took effect in July 2026, includes provisions for transparency and user awareness, such as informing users they are interacting with a machine.

China revises export controls for AI technologies

On October 16, the Ministry of Commerce and the Ministry of Science and Technology proposed changes to China's export control catalog, including restrictions on 23 AI-related technologies like large models and computing clusters. Open-source and publicly published technologies are exempt from these restrictions, with the consultation period ending on November 15, 2026.