An intelligence service, now a regular columnist on WeChat. The Chinese Ministry of State Security (MSS), the country's main espionage and counter-espionage agency, warns that accelerated progress in artificial intelligence can backfire on China: leaks of sensitive data, manipulation of models, tailor-made influence operations.
The warning comes from a country which claims the world podium in AI and which has made its open models an instrument of influence. Massively disseminating a technology while blocking its uses is a balancing act.
Key Points
MSS, China's main intelligence agency, describes AI as a vector for data leaks and influence operations
Anti-espionage law expanded in 2023, mandatory labeling of content generated by AI since September 2025: the legal arsenal is tightening
Beijing applies the doctrine tested on cryptos to models, between banning open networks and promoting the traceable digital yuan
According to several analyses, China still retains 194,000 BTC from the PlusToken seizure, a point now disputed
AI, a new front for Chinese national security
Since 2023, the Ministry of State Security has regularly published educational posts on WeChat, a rare exercise for an administration accustomed to discretion. The themes rotate in a loop: cartographic data, connected objects, electric vehicles, weather stations, and now language models. The scenarios described come back to the same fears: civil servants feeding chatbots with internal documents, or cloned voices used for social engineering. Another risk documented by the MSS: poisoned datasets before training the models.
The legal arsenal followed suit. The anti-espionage law revised in 2023 extended the notion of state secret to any document or data affecting national security, a formulation broad enough to include training bases. Interim measures on generative AI govern general public services, and mandatory labeling of content produced by AI has applied since September 2025.
The official doctrine is contained in three adjectives. Before the Politburo meeting in April 2025 for a study session devoted to artificial intelligence, Xi Jinping set the course.
“Artificial intelligence must be safe, reliable and controllable.”
Xi Jinping, President of the People's Republic of China
Three months earlier, the release of DeepSeek's R1 model had caused American technology stocks to plummet, with nearly $600 billion in capitalization evaporated in one session for Nvidia alone. The Chinese state apparatus celebrates this industrial breakthrough and at the same time describes the models as a dream gateway for opposing services.
Crypto: the same distrust, from banned bitcoin to digital yuan
Beijing has honed this reading grid on cryptocurrencies. In September 2021, ten administrations led by the central bank declared all activities related to digital assets, from trading to mining, illegal. Bitcoin has not left the country, however, over-the-counter exchanges having simply shifted to the discretion of private messaging.
The official alternative has a name: the digital yuan. Its international operations center opened in Shanghai in 2025, with the ambition of internationalizing a currency that is programmable and traceable to the nearest cent. In Hong Kong, the licensing regime for stablecoin issuers came into effect on August 1, 2025, before mainland authorities asked brokers to freeze their work on tokenizing real assets and tech giants their stablecoin projects.
The lock works as long as value is flowing over managed pipes. It crumbles as soon as it migrates to infrastructures that no one operates: decentralized calculation markets like Bittensor or Render and self-hosted wallets. An open source model redistributed outside any jurisdiction is just as immune. A model published under a permissive license escapes its author from the first download, and this is exactly the risk that the MSS points out.
