China's spy chief says AI can strike at Communist Party rule

Chen Yixin, minister of state security, published an article on Sunday in China Cyberspace, the journal of the Cyberspace Administration of China. He wrote that artificial intelligence is penetrating society at a speed, breadth and depth that is reshaping economic structure, power, innovation and social governance. He then said abusive use of the technology by adversaries could directly threaten China's political security, institutional security and ideological security.

That sentence is the piece. Western safety debates this month have circled loss of control and harm to humans. Chen's list is different. He put the Communist Party's hold on power first, then cyber defence, then military competition with the United States.

He named specific US systems. The New York Times reported that he cited Claude Mythos and GPT-5.5-Cyber as examples of new threats to Chinese networks. He said foreign intelligence services might use models for large-scale espionage and for attacks on critical infrastructure. He also warned that people with ulterior motives could generate fake video and audio and wage a propaganda war against China.

A security ministry text, not a science paper

Chen called AI a new arena for strategic rivalry among major powers. He asked for more party control over the sector, stricter government oversight, stronger risk-prevention frameworks and a larger Chinese role in global AI rules. The same weekend, at the BRICS summit in New Delhi, Xi Jinping proposed an AI-powered initiative on industrialisation and supply chains. The two messages travel together. Beijing wants to write governance language abroad while tightening the domestic grip.

The contrast with Dario Amodei's Saturday essay is sharp. Amodei wants labs and governments to slow capability growth because models may soon improve themselves faster than people can check them. Chen wants the party to own the stack so that foreign models and foreign services cannot use the same tools against the regime. One text fears an uncontrolled machine. The other fears a controlled machine in the wrong hands.

Neither text is a statute. Chen's article does not list new criminal penalties or a licensing timetable. It does tell every ministry and every company that political security is now an official AI risk category, written by the man who runs the civilian intelligence service.

What follows from naming US models

Once a security minister puts product names in a party journal, procurement officers and platform censors have a signal. Domestic champions gain a reason to argue that foreign weights should stay off Chinese networks. US firms already face export controls on chips. They may now face a political argument that their models are instruments of espionage even when they are sold as chat products.

Chen also folded ordinary privacy and commercial secrets into the same frame. State data, business secrets and personal privacy are at stake, the South China Morning Post quoted him as saying. That widens the brief from ideology to the entire data economy, which is how security agencies usually grow a mandate.

Readers outside China should not treat the article as proof of a particular plot. It is proof of a priority. The highest intelligence official in the country has said, in a state journal, that AI can be used to loosen the party's grip, and that the answer is more party control. How that instruction is written into rules for model licensing, data export and cross-border APIs is the next document to watch.