🧠 Food for thought

Implications, context, and why it matters.

China’s agent policy sets a detailed rulebook for high-risk AI

• The guidelines set a tiered system across 19 sectors 1. Healthcare, finance, and transportation are in the highest-risk group. Judicial services plus public security are there too 1.
• Firms in those fields must file AI agent products, pass compliance checks, and keep recall systems ready 1.
• The rules limit autonomy. Agents “may not act beyond the scope of a user’s explicit authorisation,” while users must be told about autonomous actions and retain the right to override them 1.
• They also ban manipulative conduct, including emotional dependency or addictive use, with extra safeguards for minors plus older adults 1.

The framework sits within China’s wider AI drive

• The policy follows the country’s “AI Plus” effort to spread AI into daily industry use, echoed in the 2026 government work report, which calls for expanding the initiative and speeding up intelligent agent development 2.
• It also fits Beijing’s interest in “embodied AI” (AI built into physical machines such as robots) for robotics and factory automation, though these guidelines do not describe an embodied-AI policy 3.
• National direction can lead to local programs. Chongqing offers subsidies of up to 5 million yuan (US$736,000) for companies building “Future Factories” 4. It also backs industrial intelligent agent research, development, plus testing platforms 4.
• The rules also propose an Agent Interoperability Protocol (AIP), a standard for AI agents to work together 1. They also outline an “Intelligent Internet” with agent registration plus digital identity 1. It would include capability declarations, coordination protocols, plus compliant payment systems 1.