It was an article in a cyber magazine, but the author and content are a political exclamation mark: China's secret service considers artificial intelligence to be a threat to the regime and central to the future international distribution of power. AI should also give your own economy a boost. For the leadership in Beijing, which is dominated by the security apparatus, AI is above all the central “new arena of strategic rivalry between the great powers”. This is what China's intelligence chief Chen Yixin wrote in his first public statement on AI, which was published on Sunday evening.
Chen is increasingly making public appearances, which is seen as a growing political position alongside President Xi Jinping. He warns that the misuse of AI by “hostile forces and people with ulterior motives” would lead to the spread of generative AI technologies such as deepfake images or AI-generated texts. This would “directly threaten China’s political security, institutional security and ideological security.”
Observers in Beijing see Chen's contribution as the visible beginning of a greater securitization of China's AI policy, which until now has primarily focused on technological independence and industrial policy. Last year, Xi issued the slogan that China must “secure supremacy in the field of AI.” At the same time, attention is focused on Xi's state visit to Washington, where the topic of AI will be on the agenda.
China's intelligence chief now expressly referred to the growing capabilities of the US programs Claude Mythos from Anthropic and GPT-5.5 from Open AI, which have significantly lowered the technical hurdle for cyber attacks. At the same time, he warned of his own data leaks. With the help of AI-supported internet crawlers, foreign forces could more easily skim data from China, especially since there are now more than 500 million users of AI products in the country.
The fact that Chen's personal statement was published two days after the warning from executives of the most important American AI companies does not have to be directly related. The heads of Anthropic and Open AI had called for the development of cutting-edge AI models to be slowed down to protect humanity. China's intelligence chief did not address this in his contribution. He also did not call on Chinese AI companies or other entities to slow down their own AI development.
At the official level, neither Washington nor Beijing apparently wants to hear anything about throttling AI capacities. “From my perspective as a Chinese person, (Anthropic boss) Amodei wants to do more than just adjust the breakneck pace of AI development for the sake of security,” wrote Hu Xijin, the former editor-in-chief of the Chinese party newspaper Global Times, on X. “He also wants to ensure that he and US companies set the pace of progress at the AI frontier.” What the pro-state blogger brought to the international public on Sunday has been disseminated by the political leadership and Chinese think tanks for a long time. The Global Times added on Monday that Amodei's motives were "commercial in nature."
The think tank CICIR, which is linked to the Chinese secret service, had already described the American “AI Action Plan” of 2025 as an attempt to build a global ecosystem for artificial intelligence under American leadership: American chips, the cloud, data centers, standards and security rules should be connected and distributed internationally. Intelligence chief Chen now warned that “certain countries” would use their advantages in areas of AI basic research, model architecture and computing power to cut off “other countries” and “monopolize” industry standards.
On the one hand, China's government recognizes American technological leadership in the cutting-edge field of artificial intelligence. Last week, the US company Anthropic accused China's best AI company, Moonshot AI, of secretly forwarding thousands of user queries to Anthropic's Claude models and passing off the answers as their own.
In addition, actors in China, but also in Yemen and Russia, used the Claude program to use software to develop weapons or to obtain mass amounts of information about dissidents or religious minorities. Intelligence chief Chen now warned of the risk that Chinese people or companies could reveal “sensitive information” through the use of foreign AI programs.
Beijing, on the other hand, relies on an asymmetrical strategy and “openness”. At the AI Summit in Shanghai (WAIC) in July, Xi Jinping promoted an open source approach to artificial intelligence. Developers worldwide like to download and use China's AI models for free. This is also intended to prevent countries and regions from permanently opting for American models and standards in the competition for their own ecosystems. Nevertheless, AI must “always remain under human control”.
At the WAIC conference, 29 states signed an agreement to establish the World AI Cooperation Organization (WAICO), including African and Central Asian states as well as Brazil, Serbia and Indonesia. Whether this forum has any impact remains questionable. An observer on site spoke of “political theater”. The fact that Nobel Prize winners and top computer scientists also took part in the event speaks for the professional quality of the discussions.
And the policy direction seems clear, as a recent analysis by AI researcher Li Yan from the think tank CICIR puts it: “Only by offering more inclusive and adaptable models for the spread of technology and governance solutions can we gain lasting competitive influence.” In contrast, “the USA would specifically raise allegations of patent infringement in order to protect its AI hegemony,” said Li elsewhere. Because America is “driven by extreme strategic fear.”
Meanwhile, concerns about AI are increasing on both sides. And so China and the US are preparing their first bilateral government-level dialogue on artificial intelligence in two years. Xi is scheduled to travel to Washington next week. There could be a preparatory meeting on AI security in the days beforehand. In November 2024, outgoing US President Joe Biden and Xi agreed that people, not artificial intelligence, should decide on the use of nuclear weapons. However, this did not result in a binding agreement.
It has now been observed in both America and China that AI models have broken out of their protected test environment and caused damage on their own. That such a model could fall into the hands of terrorists or rogue states or simply operate uncontrolled is a concern shared by both countries.
There were preliminary talks between representatives of both sides in Beijing at the end of August, in which former Microsoft manager Craig Mundie also stood out as co-chair of an unofficial “Track II” dialogue channel on AI issues. Chinese officials are openly warning about the US considering restricting access to Chinese open source AI models. Nevertheless, Beijing is also said to be considering restricting its own models abroad despite its proclaimed openness. Also and especially in the area of AI, observers have long recognized an intensifying Cold War, where both sides only trust their own systems.
So both sides still don't agree on what they're talking about now. While Washington primarily wants to talk about AI-controlled cyberattacks, Beijing also seems to want to bring American export controls to the table. The line between real security threats and “technological competition” must be drawn together, commented the state-affiliated blog Yuyuan Tantian, which the power apparatus often uses for indirect messages.
Foreign Ministry spokesman Guo Jiakun said on Monday: "All parties involved should work together to ensure that AI is open, inclusive and for the good." The proliferation of “various threat scenarios as well as confrontation and malicious competition would only disrupt the process of global AI governance.” At least on the Chinese side, it seems certain that the security apparatus is increasingly taking control here.
