China promotes human-centered AI governance

China on Tuesday released a technical guideline on the ethics and safety of artificial intelligence applications, offering a Chinese approach that emphasizes human-centered development, AI for good and human leadership over AI systems.

The document, titled Ethics-Safety Guidelines for Artificial Intelligence Applications 1.0, was released at a sub-forum of the 2026 China Internet Civilization Conference. The annual conference is being held in Nanning, capital of the Guangxi Zhuang autonomous region.

Released by the National Technical Committee 260 on Cybersecurity, China's key body responsible for cybersecurity standardization, the guideline is described as a "principle-based and reference technical document". It covers AI application development, service provision and use, and provides guidance for organizations and individuals engaged in AI-related activities.

It outlines major principles including reasonably controlling risks, maintaining openness and transparency, protecting privacy and security, ensuring controllability and trustworthiness, agile co-governance, and inclusive sharing.

The guideline says AI applications should always serve "the common well-being of humanity", calling for ensuring that "the leadership over AI applications belongs to humans", with human control, emergency response and intervention mechanisms established at key stages.

The document also urges relevant parties to retain "human judgment, supervision, intervention and correction" at key stages to avoid excessive replacement of human decision-making by AI.

For users, it advises moderate AI use and a proper understanding of AI emotional services. It says AI should be treated as "a tool to assist real life" and urges users to avoid excessive dependence, addiction, or replacing real interpersonal communication and activities with AI.

For information and communication services, the guideline calls for stronger risk governance of AI-generated and recommended content, warning against "information cocoons, cognitive misleading and cognitive degradation".

The guideline also encourages building an open-source innovation ecosystem, including open access to AI models, tool components and evaluation benchmarks, while improving the security capacity of the open-source ecosystem.