China's CAC, NDRC and MIIT issue joint guidelines on AI agents

China's Cyberspace Administration of China (CAC), National Development and Reform Commission (NDRC) and Ministry of Industry and Information Technology (MIIT) jointly issued implementation guidelines on AI agents on May 8, 2026. The guidelines set safety and controllability as a foundational requirement for how AI agents are built and used across the country.

What it is

The document defines AI agents as systems that can act with limited human direction rather than as conventional chatbots. The guidelines describe them as intelligent systems with multiple autonomous capabilities operating alongside both digital and physical infrastructure. The guidelines organize measures into four areas covering technical foundations, safety and security, application-driven rollout across 19 scenarios, and ecosystem building.

What it requires

The guidelines direct developers to strengthen their capacity to detect, intervene in, halt and recover from agent behavior that departs from intended operation. Developers are also told to guard against technical risks tied to agent operation, including data poisoning and manipulation of underlying algorithms. Users must be kept informed of an agent's autonomous decisions and must keep final decision-making authority over outcomes. On infrastructure, the guidelines call for consolidating technical foundations and establishing shared standards and protocols before agents scale into wider use.

What security leaders should do

Security leaders operating agentic AI systems that touch Chinese-market data, users or infrastructure should inventory which agents carry autonomous decision authority and document the scope of actions each one can take without a human check. They should map existing containment controls, including behavior monitoring and human approval steps, against the guidelines' emphasis on detecting and reversing improper agent behavior and on preserving user final decision authority. Because the guidelines were issued jointly by cybersecurity, industrial policy and technical-standards regulators, compliance teams should expect related technical standards and sector-specific rules to follow and should track CAC, NDRC and MIIT channels for implementing details rather than treating the guidelines as a final, static requirement.