As the deployment of autonomous AI agents accelerates across China’s digital economy, the country’s top internet regulator has stepped in to establish clear boundaries. The Cyberspace Administration of China (CAC) has published draft regulations specifically targeting agentic AI, with a core mandate: humans must remain in the loop.
According to a report by The Register, the draft rules represent one of the world’s first comprehensive attempts to govern AI systems that can execute tasks, make decisions, and interact with other software autonomously. The regulations reflect Beijing’s dual desire to foster innovation in the booming agentic AI sector while maintaining strict control over its societal and economic impacts.
The Mandate for Human Oversight
The central pillar of the draft regulations is the requirement for human oversight. The CAC stipulates that developers and operators of AI agents must “clarify the reasonable boundaries and required authority for various decision-making methods, such as decisions limited to the user, decisions requiring user authorization, and autonomous decisions by the intelligent agent.”
Crucially, the draft requires that developers “ensure that users have the right to know and the final decision-making power regarding the autonomous decisions made by the intelligent agent and that the intelligent agent’s actions do not exceed the scope authorized by the user.” In other words, an AI agent may act autonomously within defined parameters but the user must always be able to see what it is doing and retain the ability to override it.
This is particularly relevant for agents deployed in high-stakes environments. The CAC’s draft identifies a range of tasks that AI agents are expected to take on: marking homework, analyzing medical images, evaluating employee performance and recommending promotions, helping with disaster relief efforts, and managing “the entire bidding and tendering process.” These are not abstract hypotheticals. They describe deployments already underway or imminent in China’s rapidly expanding AI ecosystem.
Standards, Datasets, and International Participation
Beyond the human oversight requirement, the draft calls for the development of datasets that accelerate the responsible development of agentic AI, along with security standards. The CAC also calls for mandatory behavioral standards for AI agents “in fields such as healthcare, transportation, media, and public safety,” sectors where the consequences of autonomous errors are most severe.
Notably, the draft expresses China’s intention to participate in international fora that develop global standards for agentic AI. This signals Beijing’s ambition to be a rule-maker, not merely a rule-taker, in the emerging global governance architecture for autonomous AI systems. As Singapore released the world’s first formal governance framework for agentic AI, a principles-based document that Chinese companies have been closely studying, China’s CAC is opting for binding, enforceable rules that go further in specifying compliance requirements.
Balancing Innovation and Control
The introduction of these draft regulations comes at a critical time for China’s AI industry. Companies like Alibaba, Baidu, and Tencent are heavily investing in agentic capabilities, integrating them into everything from enterprise software to e-commerce platforms. The agentic AI boom has already driven token consumption to 140 trillion per day, underscoring the scale and speed of deployment.
China’s approach to agentic AI governance is notably more prescriptive than the frameworks being developed in the West. While the US and EU are still debating principles-based approaches, the CAC is moving toward binding, enforceable rules. This reflects a broader pattern in Chinese AI governance: the state prefers hard rules over soft guidance, and it prefers to act quickly rather than wait for international consensus.
This approach has both advantages and disadvantages. On the positive side, it provides industry with clear, actionable compliance requirements. On the negative side, it risks locking in specific technical requirements that may become obsolete as the technology evolves. The requirement that users retain “final decision-making power” over autonomous actions, for example, may be difficult to implement in practice for agents that execute hundreds of micro-decisions per session.
A Regulatory Signal at a Diplomatic Moment
The timing of the CAC’s draft regulations is significant in the context of the upcoming Trump-Xi summit on May 14–15. By publishing a comprehensive agentic AI governance framework, Beijing is reinforcing its position as a responsible AI power — one that takes safety seriously and is willing to impose binding rules on its own industry.
This narrative is useful diplomatic currency at a summit where AI governance is expected to be a key discussion point. The draft regulations also connect directly to the military AI agenda at the summit: the principle of human oversight over autonomous decision-making is as relevant to military AI as it is to commercial applications. China’s public commitment to this principle in the civilian domain may be used to reinforce its stated position that “human primacy must be upheld” in military AI as well.
Whether the US will view China’s regulatory approach as genuine progress or as a strategic communications exercise is another question entirely. The draft regulations are currently open for public consultation, and the final rules may look quite different from the current draft. But as a statement of intent, they represent a significant step in China’s effort to shape the global governance of agentic AI on its own terms.
(See also: China Unveils National Guidelines to Regulate and Promote AI Agents)