China steps up safeguards against AI losing human control

Chinese developers have been promoting open-weight AI models, allowing parameters to be downloaded, inspected, and modified, unlike US rivals Anthropic and OpenAI. This approach contrasts with the US companies' private stance on model specifications. Chinese policymakers have been preparing for risks associated with increasingly powerful AI models, which they view as manageable rather than existential threats.

While concerns in the US focus on potential existential threats, China emphasizes the need for technical standards, regulations, and state oversight to contain risks. China's approach differs from US experts, who advocate for independent monitors in AI companies and stronger safeguards. Chinese regulations include provisions for developers to improve their ability to detect, intervene in, block, and recover from improper agent behavior, as well as guard against risks such as data poisoning, algorithm manipulation, and system vulnerabilities.

The country is also developing a mandatory national standard for AI agent safety, which would be the world's first of its kind. Chinese officials have expressed concerns over US models like Anthropic's Mythos and OpenAI's GPT-5.5-Cyber, warning they could pose risks to China's critical information infrastructure. However, Chinese developers promote open-weight models partly because cybersecurity teams can inspect, modify, and deploy them for defensive work.