China outlines ethical governance framework for embodied AI

China has released a report outlining an ethical governance framework for embodied AI, addressing risks associated with systems that can perceive their surroundings, make decisions and act in the physical world.

The Research Report on Ethical Governance of Embodied Intelligence was released on 13 October by China’s National AI Standardization General Group, according to Chinese media reports carried by CCTV News and reproduced on the official Digital China website.

Embodied intelligence refers to AI systems that interact with the physical environment. Applications include robotics, industrial automation and other systems capable of acting in the real world.

The report treats the ethics of embodied intelligence as part of broader AI ethics, while identifying risks arising from the direct interaction between intelligent systems, people and their surroundings.

The report identifies six categories of ethical risk: potential harm to human autonomy, social relationships and development rights; physical injury and threats to public safety; privacy violations and misuse of personal information; bias, discrimination and exclusion from services; identity misidentification and overestimation of system capabilities; and unclear accountability, including difficulties reconstructing incidents.

The framework classifies risks into five levels, designated R1 to R5, according to the severity of potential consequences.

It also sets out six basic governance principles: human-centred development, safety and controllability, privacy protection, fairness and inclusion, transparency and explainability, and accountability and traceability.

Together, these principles address the potential consequences of embodied AI systems’ actions and the responsibilities of those developing, deploying and operating them.

The report calls for ethical requirements to be incorporated throughout an embodied AI product’s lifecycle, covering design, research and development, deployment, operation and servicing, and decommissioning.

Each stage should have defined protection requirements and clearly assigned responsibilities, according to Zhang Shizong, deputy director of the Information Technology Research Centre at the China Electronics Standardization Institute (CESI).

Zhang outlined three areas in which safeguards should operate.

The first is physical safety, through limits on what a machine can do. The second is data protection, through controls on what a system can collect through its sensors. The third is preserving human choice, including making clear that users are interacting with a machine and providing ways to switch it off.

These measures are intended to address risks associated with systems that can affect people’s physical safety or collect information about individuals and their surroundings.

The report also highlights the importance of reconstructing incidents and establishing accountability, which can involve multiple parties responsible for a system’s design, development, deployment and operation.

Zhang said China would accelerate work on national standards, including guidelines for the ethical governance of AI and embodied intelligence.

A separate entry on China’s National Standard Information Public Service Platform lists a national standards project titled Artificial Intelligence — Guidelines for Embodied Artificial Intelligence Ethical Governance.

The project, numbered 20263039-Z-469, was initiated on 28 May 2026 and has a planned 12-month development period. The platform lists Peking University, CESI, Peking University Wuhan Institute for Artificial Intelligence and the Beijing Key Laboratory of Artificial Intelligence Safety and Standards among the drafting organisations. Its recorded status is “drafting”.

The standards project is separate from the research report released in October. Publication of the report does not, by itself, establish that the proposed national guidelines have been finalised or adopted.

The project sits alongside other planned standards for embodied AI, covering areas such as evaluation, training-data infrastructure, system requirements and the classification of ethical governance scenarios.

China has also been developing broader guidance for AI ethics and safety. In May 2026, the National Cybersecurity Standardisation Technical Committee published the AI Application Ethical Security Guidelines 1.0. An accompanying explanation published by the Cyberspace Administration of China described provisions addressing human control, fairness, transparency, privacy protection, risk monitoring and incident traceability.

That guidance concerns AI applications more broadly, whereas the October report focuses specifically on embodied intelligence and the ethical questions raised by systems operating in the physical world.

The proposed national standard remains in development. Further details on how the principles outlined in the report may translate into specific technical requirements and operational safeguards will depend on the standards-development process.