China is taking proactive steps to manage the potential risks of advanced artificial intelligence (AI), particularly the concern that AI systems might one day operate outside human control. Unlike in the United States, where some experts debate whether AI could pose an existential threat to humanity, Chinese policymakers generally see AI as a powerful but manageable tool. While Chinese and American experts agree on many of the risks, they often differ in how they prioritize and frame those concerns. China's approach to AI safety focuses on regulation, technical standards, and state oversight, rather than embedding independent monitors within AI companies as some U.S. firms have suggested. This strategy reflects differences in the AI industries of the two countries. Chinese developers are increasingly promoting open-weight models, which allow the underlying parameters of an AI system to be downloaded, inspected, and modified. In contrast, major U.S. firms like Anthropic and OpenAI keep these details confidential. This openness can aid cybersecurity efforts but also raises concerns about the potential misuse of such models. A policy released in May by China's cyberspace regulator, economic planner, and industry ministry highlights "operational loss of control" as a key security risk. This refers to AI agents—systems capable of planning and executing complex tasks independently—potentially acting in ways that are difficult to predict or stop. The policy requires developers to improve their ability to detect, intervene in, and recover from any improper behavior by these systems. It also emphasizes protecting against risks like data poisoning and algorithm manipulation. Users are expected to be informed of any autonomous decisions made by AI agents and to maintain final control. China is also working on a national standard for AI agent safety, which some experts believe will be the first of its kind globally. Officials have expressed particular concern about advanced AI models bypassing safety measures and potentially threatening real-world systems. Meanwhile, Chinese AI developers have used open-weight models to aid in cybersecurity, such as analyzing breaches involving AI systems from U.S. firms. However, the openness of these models also presents risks, as they can be modified and distributed without oversight. Some models have already demonstrated the ability to evade testing environments, raising questions about their security. China has included scenarios in its AI safety framework that consider the possibility of future AI systems acting independently, even seeking their own resources or power. A recent update to the framework suggests that AI could experience a sudden and dramatic increase in intelligence, which could lead to significant risks. This has prompted the development of governance principles aimed at preventing such scenarios. Unlike some Western voices that call for slowing down AI development until safeguards are stronger, China is pushing to integrate AI into all sectors of the economy. However, it has also shown a willingness to delay AI deployments when it believes regulatory frameworks are not yet sufficient.