The framework centers on safe, controllable, reliable, and trustworthy deployment with human oversight and clear decision boundaries, specifying which actions require human control or authorization and which can be autonomous.

It envisions an Intelligent Internet built on agent registration platforms, digital identities, capability declarations, interoperability protocols, and widespread IPv6 use to enable cross-agent communication and governance.

China unveiled a comprehensive AI agent regulatory framework aiming to guide adoption across major industries, with a target of 70% usage by 2027.

The program spans broad industrial and societal sectors—from manufacturing and energy to finance, healthcare, education, transportation, and government services—integrating agents into the real economy and governance structures.

There is an emphasis on third-party testing, certification, and industry self-regulation to ensure functionality, safety, and compliance of agent products.

Regulators recognize agents’ advanced capabilities—autonomous perception, long-term memory, tool use, cross-platform task execution, and multi-agent coordination—as more than traditional chatbots.

The framework supports indigenous controllability, open-source ecosystems, domestic OS/chips, and active participation in international standards development to shape global agent governance.

Authorities approved 19 specific use cases to spur deployment and state procurement, covering smart manufacturing, financial risk control, scientific research, and embodied robotics.

The document envisions future benefits in smart governance, public administration, social management, and industrial optimization, while aiming to prevent systemic risks and preserve governance capacity.

A tiered governance model proposes high-risk sectors—such as healthcare and public safety—face mandatory standards, filing, recalls, and oversight, whereas lower-risk applications rely on self-regulation and platform governance.

The framework outlines a tiered system where high-risk domains require government filing, product recalls, and stringent standards, while low-risk areas depend on self-regulation.

China’s approach differs from the U.S. by emphasizing practical deployment and governance alongside industrial development, rather than adopting a pause-first stance on catastrophic AI risk.