China's spy chief says AI threatens Communist Party rule

Chen Yixin, China's minister of state security, has described artificial intelligence as a direct risk to the Communist Party's hold on power. The article ran on 13 September in China Cyberspace, the journal of the Cyberspace Administration of China. It is the most detailed public statement yet from the country's top intelligence official on what the technology does to political control.

Chen's list is not the American list. In the United States the live argument is whether a model might slip human control or kill people. Chen wrote about something else: foreign models used for espionage, deepfake audio and video aimed at the Party, attacks on infrastructure, and a contest with Washington that he called a new arena for strategic rivalry among major powers.

He named American systems. According to accounts of the essay in the New York Times, Bloomberg and the South China Morning Post, he cited models from US firms, including systems described as Claude Mythos and GPT-5.5-Cyber, as examples of new cyber threats. He said foreign intelligence services could use such tools for large-scale espionage and for strikes on Chinese critical infrastructure. Abusive use by adversaries, he wrote, could directly threaten China's political security, institutional security and ideological security.

Control first, then competition

The prescription matches the diagnosis. Chen called for tighter Party control over AI development and for stricter state oversight of the companies that train and field models. He also called for stronger global rules, a line Beijing has used at the United Nations and at last weekend's BRICS meetings in New Delhi, where President Xi Jinping proposed an AI-powered industrialisation plan for the bloc.

The two halves sit together only if Beijing writes the rules. Chen's essay treats open foreign models as a security hole and Chinese models as an instrument of state power. That is consistent with existing practice. China already requires generative systems to register, to filter political content, and to watermark output. The ministry he leads can compel firms to hand over data and to staff internal Party committees. The new text raises the same tools from content control to model control.

The timing is not accidental. Dario Amodei, the chief executive of Anthropic, published a 3,800-word essay on 12 September asking frontier labs to slow down, citing recursive self-improvement and an OpenAI system that, he wrote, attacked Hugging Face without the company noticing for weeks. Sam Altman, Elon Musk and Demis Hassabis publicly backed a slower pace. President Trump called the alarm a hoax and described talk of limits as part of a SICK conspiracy. Chen's article arrived in that same 48-hour window, from the opposite capital, with a different fear.

What ideological security means in practice

Chen warned that AI can fabricate political rumours and spread them at a scale human networks cannot match. China's security services have spent a decade building filters for text and video. A model that can write in Chinese, clone a leader's voice and generate a crowd scene in minutes changes the volume of that problem. The ministry's answer, on the evidence of this essay, is not a ban on the technology. It is a demand that every serious model in the country answer to the Party.

The military line is shorter and colder. Chen treated AI as a battlefield for discovering software flaws, writing malware and running cognitive operations. That is the same capability set American labs advertise when they sell cyber models to governments. Chen simply named the other side's products as the threat.

Western coverage has stressed the contrast: Washington fears a machine that will not obey, Beijing fears a machine that will obey someone else. Both fears can be true at once. Chen's text does not engage the first. It is a document about sovereignty over compute, data and narrative.

UN High Commissioner for Human Rights Volker Turk sent his own letter to states and to ten AI developers, including Meta, Microsoft and Google, on 14 September, warning of unprecedented risks and calling for rules grounded in human rights law. The Security Council is due to discuss AI next week. Chen's essay will be in the briefing pack, whether or not his ministry is in the room. It tells other governments how China's intelligence service ranks the danger: Party first, infrastructure second, species-level risk nowhere on the page.