Key facts
- China's state security minister Chen Yixin warned that advanced U.S. AI models pose risks to critical information infrastructure.
- China's AI safety framework includes scenarios where AI might autonomously obtain resources, replicate itself, and seek power.
- President Xi Jinping stated that AI should 'always remain under human control'.
- China issued guidelines for AI agents requiring developers to manage improper behavior and retain user decision-making authority.
- Chinese AI developers promote open-weight models, while experts note risks of modification and redistribution.
- The Cyberspace Administration of China (CAC) has developed AI safety frameworks since September 2024.
China's state security minister, Chen Yixin, has warned that advanced artificial intelligence models, particularly closed-source U.S. systems, pose significant risks to the nation's critical information infrastructure and overall security. These concerns are being integrated into China's governance strategies for AI development.
Chinese policymakers view the potential for advanced AI to escape human control as a serious threat, prompting the development of safety frameworks and regulations. The Cyberspace Administration of China (CAC) first outlined loss-of-control scenarios in September 2024, with an expanded version in September 2025 detailing risks such as AI autonomously acquiring resources, replicating itself, and seeking power. A key governance principle introduced is 'trusted application, preventing loss of control'.
President Xi Jinping has emphasized that AI must "always remain under human control," a sentiment echoed by officials advocating for cautious approaches to military AI to prevent strategic miscalculation. China is accelerating research into broader AI legislation and has issued specific guidelines for AI agents, which act more autonomously than chatbots. These guidelines require developers to improve their ability to manage improper agent behavior and ensure users retain final decision-making authority.
While Chinese developers promote open-weight models, citing cybersecurity benefits, experts acknowledge the risks associated with their modification and redistribution. The country is also accelerating research into broader AI legislation.