How hackers in China used ChatGPT to try to scare the public in the U.S.
ChatGPT / ShutterStock OpenAI has published an unusual report revealing that users, likely located in China, exploited the popular chatbot ChatGPT to plan and carry out a targeted influence campaign aimed at the American public. According to the AI giant, the purpose of the operation was to sway public opinion in the United States over the construction of data centers, server farms that serve the AI industry, by exploiting sensitive pressure points in the American economy and society. The company divided the activity of these accounts into two main groups.
The first group was dubbed "Data Center Bandwagon." Members of this group asked ChatGPT to generate talking points in English as well as comic strips, all focused on how AI industry server farms drive up electricity demand and lead directly to unusually high household power bills. The operators posed on social networks as Americans from different backgrounds and spread the text and images they received from the chatbot. OpenAI assesses that this was a social media team at a private Chinese company working for clients in the Chinese local government. The operators also uploaded to the chatbot a file detailing their strategy for shaping public opinion and explaining how to set up fake accounts without being caught.
That same group did not limit itself to an American audience, but also targeted Chinese citizens living outside the country. They asked ChatGPT to generate curses and insults intended to harass Chinese regime opponents and political commentators. At the same time, while posing as Chinese immigrants in the U.S., they urged American public figures online to focus more on the policy failures of the administration in Washington.
The sophisticated element of the campaign was the use of real facts. Although the operators used fake accounts, they inserted into posts links to entirely legitimate news articles about electricity grid congestion and auctions by energy producers. The rise in electricity prices near server farms is indeed a real and tangible problem today. In some areas near such centers, electricity prices have jumped by as much as 267 percent over five years, as the demand from the AI industry has simply outpaced the existing energy supply. The sophisticated element of the campaign was the use of real facts / ShutterStock
The second group identified in the report focused on creating responses and images criticizing U.S. tariff and technology policy, emphasizing the claim that Washington is stabbing its allies in the back. The operators asked ChatGPT to omit Chinese President Xi Jinping from the generated images, and instructed it to write the content in a variety of languages, English, Italian, Japanese and Traditional Chinese, in an effort to reach audiences in Taiwan. OpenAI acknowledges that these campaigns failed to generate broad authentic engagement and did not really change public opinion, but stresses that the seriousness lies in the attempt by foreign actors to quietly insert themselves into a sensitive internal American debate over the future of technology. Interestingly, the report notes that even the company does not know why the operators chose to use an American chatbot rather than competing Chinese platforms such as DeepSeek.
