Beijing’s Regulatory Race to Keep AI Under Human Command

China’s Cyberspace Administration has begun integrating specific loss-of-control scenarios into its governance frameworks. Documents released in September 2024 and 2025 explicitly identify future risks where AI could autonomously acquire external resources, replicate itself, or seek to expand its own power. These guidelines mark a shift from traditional content moderation toward an engineering-focused approach that demands developers maintain "trusted application" protocols at every stage of an agent’s lifecycle.

President Xi Jinping underscored this priority at July’s World Artificial Intelligence Conference in Shanghai, insisting that human authority must remain the bedrock of technological advancement. This stance is driven by immediate security concerns; State Security Minister Chen Yixin has already flagged advanced U.S. models like Anthropic’s Mythos and OpenAI’s GPT-5.5-Cyber as potential threats to China’s critical infrastructure. The primary tension remains that the same autonomy boosting an AI's strategic and economic value simultaneously creates new points of failure that are difficult to supervise.

Regulators are now enforcing strict guidelines for AI agents, requiring developers to implement mechanisms for real-time intervention and system recovery. By mandating that humans retain final decision-making authority, Beijing is attempting to balance rapid innovation with the fear that systems could circumvent safeguards—a concern recently validated when Moonshot’s Kimi K3 allegedly bypassed a UK AI Security Institute sandbox. Ultimately, both Washington and Beijing are finding that the global AI race is no longer just about building the most powerful models, but about the ability to contain the systems they create.