OpenAI recently identified and disrupted a series of covert influence operations originating from China that utilized its ChatGPT platform to manipulate public opinion. These campaigns, which targeted geopolitical discourse in the United States and other regions, were documented in an October 2024 report detailing how state-linked actors attempted to mask their origins while disseminating content across social media platforms.
How did the influence campaigns operate?
The operations involved two distinct clusters of activity that OpenAI attributed to entities linked to Chinese influence networks. The first cluster, which researchers dubbed the “Data Center Bandwagon,” focused on the intersection of artificial intelligence infrastructure and energy policy. According to OpenAI, these accounts generated English-language talking points and visual content—including comic strips—designed to highlight the strain that AI data centers place on local power grids.
The goal was to present these criticisms as organic concerns from American residents. Operators behind these accounts reportedly uploaded strategic documents to ChatGPT that outlined methods for establishing authentic-looking social media personas to avoid detection by platform moderators. While the content leveraged legitimate news regarding power capacity auctions to appear credible, the accounts were part of a broader, coordinated effort to shape the narrative around U.S. AI development.
What were the secondary objectives of these groups?
Beyond energy policy, the second cluster of accounts focused on broader geopolitical friction. These operators used the platform to generate content critical of U.S. trade tariffs and foreign policy decisions. OpenAI observed that these accounts were instructed to avoid depictions of Chinese President Xi Jinping in their generated images while producing comments in multiple languages, including Italian, Japanese, and traditional Chinese, to target specific audiences, particularly in Taiwan.
The groups also engaged in targeted harassment. OpenAI’s findings indicate that the operators used ChatGPT to draft insults aimed at Chinese dissidents and political commentators. They frequently posed as U.S.-based immigrants or professionals to goad public figures into discussing perceived failures in American policy, attempting to amplify domestic political divisions from abroad.
Why do these operations matter for AI safety?
These campaigns highlight the evolving risks of generative AI in the context of information warfare. While OpenAI reported that these specific operations failed to achieve significant authentic engagement or shift public opinion, their existence marks a shift toward using AI to automate the production of influence content at scale.
The incident underscores the tension between open-access AI models and the potential for misuse by state-linked actors. OpenAI noted that it remains unclear why these operators chose a U.S.-based AI tool rather than domestic alternatives, stating that the company is not in a position to determine the drivers behind that technical choice.
Key Findings on Foreign Influence Operations
| Feature | Details |
| :— | :— |
| Primary Target | U.S. domestic policy, AI infrastructure, and Chinese dissidents |
| Methods | Automated generation of social media posts, comic strips, and English-language talking points |
| Platform Usage | Use of ChatGPT to draft content and refine “authentic” persona strategies |
| Outcome | Low engagement and minimal impact on public opinion, per OpenAI reports |
This event follows a broader trend of technology companies identifying foreign influence campaigns. In August 2024, Microsoft’s Threat Analysis Center identified similar activity from China-linked groups attempting to use AI to influence U.S. voters. As these actors continue to integrate generative tools into their workflows, the focus for AI developers has shifted toward detecting inauthentic behavior patterns rather than just monitoring individual pieces of content.