Chen Yixin calls AI a threat to Communist Party rule and names U.S. models

China’s minister of state security, Chen Yixin, published one of Beijing’s most detailed official warnings on artificial intelligence in the state journal China Cyberspace on 13 September. He described AI as a new arena for rivalry among major powers and said abusive use by adversaries could “directly threaten China’s political security, institutional security and ideological security.”

The article is not a safety paper in the American sense. U.S. executives have spent the past week arguing about models that might slip human control. Chen’s list is different. He warned that people with “ulterior motives” could use generated video and audio to run a propaganda war against the Party. He said foreign intelligence services might use AI for large-scale espionage and for attacks on critical infrastructure. He named two U.S. systems, Claude Mythos and GPT-5.5-Cyber, as examples of new cyber threats.

He also put a number on domestic spread. In June 2026, he wrote, AI products in China already had more than 500 million users. That figure is the political problem as much as the technical one. A tool that sits on hundreds of millions of phones can carry a rumour at the speed of a state media bulletin, and Chen wants the Party, not the model makers, to set the limits.

The remedies he listed are familiar to anyone who reads Cyberspace Administration texts: more Party control over the sector, tighter oversight of data, algorithms, computing infrastructure and both civil and military applications, and a “security barrier” around the technology. He accused unnamed countries of using accumulated advantages in theory, model architecture and computing power, under the pretext of national security, to block others from buying advanced equipment. The target is the United States.

On the military side Chen said AI is moving from a support role on the battlefield to a leading one, and he cited the conflict between Washington and Tehran as the current example. That sentence ties the essay to the war that has already closed the Strait of Hormuz and rerouted Saudi crude through the Red Sea. It also tells domestic readers that the ministry sees foundation models as weapons systems, not only as consumer software.

President Xi Jinping used the BRICS summit in New Delhi a few days earlier to offer an AI industrialisation plan for emerging economies. Chen’s essay is the security twin of that offer. One hand proposes shared industrial tools. The other says foreign models are an active threat to Party rule and must be fenced.

Henry Gao, a law professor at Singapore Management University, told the New York Times that Beijing will not trade data sovereignty or political security for an international safety accord. That reading fits the text. Chen wants global governance language, and he wants it on Chinese terms: the state decides what training data is allowed, which applications are high risk, and which foreign systems may not run on domestic networks.

The essay lands as Chinese firms such as DeepSeek, Moonshot, Alibaba, Tencent and Zhipu push cheaper open models despite U.S. chip limits. Chen is not asking those firms to slow down. He is asking them to accept a thicker Party layer on data, algorithms and compute. The test will be whether the next licensing round for large models in China writes Claude Mythos and GPT-5.5-Cyber into the blocked list by name, or treats them as illustrations.

U.S. and Chinese officials are due to talk ahead of a meeting on 24 September. Chen has now told those talks what Beijing’s security service thinks the subject is. It is not alignment of safety benchmarks. It is who holds the switch when a model can write malware, fake a leader’s voice, or map a power grid.