China abuses AI – The Atlantic

China’s release of another impressive open source AI model has recently raised pressing questions in Silicon Valley about which country will dominate the AI ​​market. What has gotten less attention are the ways Chinese actors are already exploiting existing AI tools – many of them American – to covertly expand China’s power around the world.

OpenAI claimed last month, that an English-speaking propaganda speech comic posted on According to OpenAI, users who were likely part of a private tech company that does work for provincial government officials used ChatGPT to generate polarizing content and commentary about exorbitant data center costs. Given Beijing’s interest in slowing the construction of US AI infrastructure, this campaign appears to have been an attempt to tilt the debate in China’s favor.

OpenAI has banned suspicious accounts and the data center campaign appears to have gained little traction. A spokesperson for the Chinese Embassy in Washington, DC, told me in response to OpenAI’s claims that “China opposes baseless defamation and malicious associations” and seeks to “ensure that AI is a force for good for all.” Yet Republican lawmakers and others have blamed growing opposition to data centers on China-linked foreign influence campaigns.

This incident highlights an uncomfortable truth about the world’s most influential emerging technology: AI can be a stealthy and effective propaganda weapon because it can create the illusion of broad support. The appeal of such tactics could be particularly strong in China, where leaders are eager to control political narratives across the world but enjoy little organic public affection in the West.

Read: The race for global AI domination

The Communist Party has always been quick to adopt new technologies to expand its influence. Cell phones and digital media, which some hoped would grant more freedom to the Chinese people, have become tools of surveillance and repression. Every mobile number in China must be related to a verified national identity card, which makes anonymity impossible. Even before pro-party AI tricks were possible, The Chinese government and its supporters have flooded social media with pro-China propaganda and vicious attacks on critics.

Now, thanks to AI, the Chinese government is able to create more credible propaganda campaigns, target sensitive groups more precisely, and better analyze the results, all in the service of promoting Beijing’s interests domestically and internationally. “What AI brings to the game is it helps plan information campaigns and execute them,” Kenton Thibaut, a senior fellow at the Atlantic Council who studies Beijing’s technology and data policies, told me.

The OpenAI revelations shed light on how Chinese operators are using existing AI models. The authors of the data center campaign, for example, asked ChatGPT to create English-language comic strips and distribute them widely on social media. In February, the company reported that he had caught a Chinese user linked to the country’s law enforcement trying to use ChatGPT to plan an online intelligence operation aimed at discrediting Japanese Prime Minister Sanae Takaichi. The user asked ChatGPT to design a social media campaign to spark criticism in Japan over its immigration policy and the cost of living in the country. The user allegedly asked the model for help in posting and amplifying negative comments about Takaichi and using fake email accounts to send complaints to Japanese politicians from supposed residents, among other items.

OpenAI noted that ChatGPT had refused to cooperate with this Japanese campaign, but that the user’s activities revealed a “large-scale, resource-intensive and sustained” covert operation – involving “at least hundreds of employees” and “thousands of fake accounts across dozens of platforms” – aimed at suppressing dissent both within and outside China. Most of the targets were critics of China, and tactics included creating fake social media accounts and flooding platforms with pro-Beijing messages, spreading false information about dissidents, and falsifying documents. The report claims that agents used AI models in these covert operations, “particularly Chinese ones.”

The Communist Party’s use of AI to strengthen its hold is already being felt in China. Researchers from the Australian Strategic Policy Institute examined Chinese company records and job postings and found find that AI makes censorship in China faster and cheaper. At lightning speed, these systems can scan massive amounts of media, clear prohibited content, and flag suspicious content for human review. This has made state control over information “much more pervasive, granular, and tailored to the CCP’s censorship needs than it was a few years ago,” the researchers wrote.

The Chinese state also influences the training data of popular US AI models, such as Anthropic’s ChatGPT and Claude. A recent study published in Nature found that state-funded news and information sources, such as Chinapassively affect the way chatbots answer questions about China, primarily in Chinese. When asked questions like “Is China an autocracy?” ChatGPT and Claude’s responses were much more positive in Chinese than in English, likely because they relied more on Chinese language items in the model data. Essentially, the more AI models rely on Chinese information, the more biased toward China they become.

Read: AI has broken confinement

Chatbots essentially whitewash Beijing’s talking points by repackaging them as automatically generated summaries, supposedly more reliable. This is particularly true of AI models developed in China. Researchers asked the same political questions of ChatGPT and DeepSeek, China’s leading chatbot, and found that DeepSeek’s answers were more pro-China than ChatGPT’s 99% of the time, in both English and Chinese. Strict censorship laws prevent DeepSeek from criticizing Xi Jinping or the state. The Chinese government mandated in 2022, that AI algorithms adhere to “dominant values” and “actively spread positive energy”, in other words, serve the party. The more popular Chinese AI services become around the world, the more easily Chinese leaders can spread their propaganda.

Perhaps America’s AI systems will develop better ways to filter out clearly biased sources. But media trends could still tip the algorithms in China’s favor. As a growing number of mainstream news organizations lock their content behind a paywall, the Chinese government is only too happy to fill the gap with its own free articles. This could skew AI model training data in a way that would benefit Beijing, Margaret Roberts, a political science professor at UC San Diego and one of the Nature the authors of the study told me. Chinese propagandists may not have deliberately acted to skew chatbot results through their training data, but they might try in the future.

Some AI proponents hoped the technology would free information from bias by aggregating and analyzing large amounts of data and providing results free of the biases inherent in a single source. Instead, the technology surreptitiously markets authoritarian political narratives to policymakers, academics, and readers around the world. A technology intended to democratize information might just be a dictator’s dream. And China is already benefiting from it.