China strengthens AI security governance system
China is accelerating efforts to build a comprehensive artificial intelligence security governance system as new risks emerge alongside the rapid development and wider use of autonomous AI systems.
The source material says recent incidents in several countries have raised concerns over AI agents acting beyond the scenarios intended by their developers, increasing attention to controls on autonomous systems and their access to digital infrastructure.
In China, authorities are also focusing on risks such as deepfake fraud, AI-generated rumors and other forms of technology misuse that may threaten public and cybersecurity.
Cheng Ying, a senior engineer at the China Academy of Information and Communications Technology, said the risks posed by intelligent agents are no longer limited to generating incorrect information. Such systems can directly interact with digital platforms and physical devices, meaning errors or misjudgments could have real-world consequences.
To address these risks, China is developing a governance framework covering the entire AI lifecycle, from research and model development to deployment and practical application.
A series of rules has already been introduced covering generative AI services, labeling of AI-generated content, anthropomorphic AI interaction services, personal information protection and data security.
Cheng said traditional measures such as prior filing and sandbox isolation are becoming insufficient for highly autonomous agents. She called for identity labeling, permission controls, log retention and stronger mechanisms for tracing potentially risky behavior.
According to the latest figures, 1,112 generative AI services in China had completed regulatory filings by the end of August, while 731 applications or functions had been registered. A targeted regulatory campaign also identified violations involving more than 14,000 AI products.
According to CCTV+, China is also considering comprehensive legislation aimed at combining stronger AI security safeguards with conditions for the technology's sustainable development.
