China has launched a plan to draft a mandatory national standard on the safety of artificial intelligence (AI) agents, as the technology moves from generating content to performing tasks and operating tools on users' behalf.

AI agents are designed to do more than answer questions or produce text. They can access data, call external tools, make multi-step decisions and perform actions across devices or cloud services. That has raised new safety concerns, including data leakage, uncontrolled permissions, tool misuse and actions that deviate from a user's intent.

The planned standard, titled General Security Requirements for Artificial Intelligence Agent Application, was listed by China's national standards information platform. It was issued on June 27, 2026, with an 18-month development cycle.

According to the platform, the mandatory standard will apply to AI agent applications deployed on smartphones, tablets, computers, wearable devices or cloud platforms, especially those able to call system or local tools to complete user requests.

Its expected technical requirements cover identity identification, system permission calls, tool use, data collection and use, human intervention for high-risk operations, input and output safety protection, log retention, dynamic monitoring, abnormal operation blocking and emergency shutdown.

The standard is being proposed by the Cyberspace Administration of China and handled by the National Information Security Standardization Technical Committee, also known as TC260. China Mobile Communications Group, the China Electronics Standardization Institute and the National Computer Network Emergency Response Technical Team/Coordination Center of China are among the main drafting organizations.

The official standards record says the mandatory requirements are intended to address personal health, life and property safety, national security, and basic needs of economic and social management.

Chinese media have described the planned standard as the world's first mandatory national standard focused on AI agent safety. Once completed, it could provide a basis for product design, deployment, operation, testing, and regulatory oversight of AI agent applications in China.