Ask a Western chatbot about China in Chinese, and it immediately spits out CCP propaganda. A new study in Nature, led by University of Oregon researcher Hannah Waight, found that Chinese state-coordinated media in AI training data pushes the models’ answers toward the Chinese government’s line. The effect is strongest when you ask in Chinese. Here’s how it happens. China ranks 178th out of 180 countries in RSF’s press freedom ranking, and it has spent years shaping what appears on its internet through propaganda and censorship. Web archives like Common Crawl scooped all of it up. AI companies then treated that archive as a neutral record of human knowledge. rsf.org/en/interview-how-chi…

A New York Times reporter expelled from China in 1991 went back for six months. The trains are faster. The surveillance is worse. Andrew Higgins counted more than 30 cameras on the 137-meter walk from his Shanghai apartment to the subway. State security officers showed up at a lawyer friend’s door and ordered him to cancel their meeting. His wife’s pen was pulled at Tiananmen security because officers thought it might contain red ink. Entry to the square required WeChat clearance, a two-hour wait, and three separate checkpoints. Men still followed him. They now use smartphones and smart glasses. AI flags the communications of foreign journalists.Posters near the temple visits told residents to report spies. The slogan was “anti-spy starts with me.” China is richer than it has ever been. It is also more certain that someone, somewhere, is an enemy. nytimes.com/2026/10/11/world…