China’s Spy Chief Frames AI Risks Around Party Control

BEIJING - China’s minister of state security, Chen Yixin, wrote in a signed article published Sunday that artificial intelligence now poses six categories of risk, led by a “systemic” threat to the country’s political security, and that the answer is the Communist Party’s overall leadership.

The essay, “Comprehensively Building a Strong AI Security Barrier and Promoting the Healthy and Orderly Development of Artificial Intelligence,” was posted on the WeChat account of China Cyberspace magazine. Chen said AI had become “the main battlefield of global technological competition” and “a new track in great-power strategic contest,” and that China must pair high-level security with efficient governance.

First among the six risks was a systemic impact on the political-security environment. Hostile forces, he wrote, were using deepfake audio and video, AI-generated images and text, and “intelligent water armies” to manufacture political rumors at low cost and in volume, conducting “public-opinion warfare” and “cognitive warfare” that “directly threaten our political security, institutional security and ideological security.” Other risks he named included industrialized cyber offense, large-scale leaks when domestic users put sensitive material into foreign AI products, monopolistic controls that fragment supply chains, strain on social governance, and a shift in warfare in which AI moves from a supporting role to a “key role,” including in the U.S.-Israel-Iran conflict.

Chen reviewed what he described as a series of important remarks by Xi Jinping on AI governance. Facing those risks, he wrote, China must “uphold the Party’s overall leadership,” strengthen systemic coordination, improve risk perception and early warning, and establish a comprehensive AI security regulatory platform.

In Chen’s telling, the technology is not merely a safety or commercial question but one of political, institutional and ideological security - to be met by tightening the Party’s hold over how models are built, watched and used.