China's spy chief says AI now threatens Communist Party rule

Chen Yixin, China's minister of state security, has given the most detailed public account yet of how Beijing's intelligence service sees artificial intelligence. In an article published on 13 September in China Cyberspace, the journal of the Cyberspace Administration of China, he wrote that AI has become "a new arena for strategic rivalry among major powers" and that hostile use of it can "directly threaten China's political security, institutional security and ideological security."

That sentence is the point. American lab chiefs spent last week arguing about whether frontier models might slip human control. Chen's list is different. He is worried about fake video and audio used against the Party, about foreign services running large-scale espionage with models, and about malware that finds its own holes in Chinese networks.

He named two Western systems. Anthropic's Claude Mythos and OpenAI's GPT-5.5-Cyber were cited, the New York Times and Hong Kong Free Press both reported, for advanced hacking ability that could put China's critical information infrastructure at risk. He also accused foreign agencies of using crawlers and data mining to lift trade secrets and personal data.

Cognitive warfare, not existential risk

Chen wrote that forces with "ulterior motives" can use generative models to fabricate political rumours and flood the public with harmful information. He called that a propaganda war and a cognitive war. He did not call for a pause in model training. He called for more Party control and tighter state oversight.

The contrast with Dario Amodei's essay last week is sharp. Amodei asked labs to slow frontier development. Sam Altman and Elon Musk said they agreed. President Trump rejected new guardrails and said a high-IQ president was enough. China's foreign ministry, separately, called Amodei's warning about Chinese models fearmongering. Chen's article shows the domestic half of that position: AI is dangerous when other people run it against the Party, and useful when the state runs it for industry and the military.

Xi Jinping used the BRICS summit in New Delhi to offer an open-source AI project as China takes the next presidency. Chen's piece landed in the same 48 hours. The two messages are meant to travel together. One is cooperative and industrial. The other is a security doctrine that treats Western chatbots as potential weapons.

What the doctrine changes

If the Ministry of State Security now treats named commercial models as infrastructure threats, procurement and network rules inside China will tighten around them. Agencies that still call Western APIs for translation or code will have a harder time explaining why. Domestic models will be asked to show that they cannot be turned into rumour mills. That is a censorship problem dressed as a cyber problem, and Chen did not hide the political end of it.

The article also sets a talking point for the Xi-Trump meeting due in Washington later this month. AI governance is already on that agenda. Chen has now given Chinese diplomats a text that says the risk is foreign cognitive warfare and cyber intrusion, not an arms race that China should slow. Officials who want export controls eased will quote the industrial half. Officials who want Western models boxed out of Chinese networks will quote Chen.

Readers who already followed the Amodei letter still get a new fact from this piece: the head of China's civilian intelligence service has put two product names on a public threat list and tied them to the survival of Party rule. That is not a lab-safety argument. It is a state-security argument, and it will shape how Chinese regulators treat foreign weights, APIs and chips in the next quarter.