OpenAI Catches China Using ChatGPT to Manipulate American Opinion on AI and Trade
OpenAI catches China using ChatGPT to manipulate American opinion on AI and trade policy
China-backed operatives used OpenAI’s ChatGPT to shape US public debate on data centers, AI infrastructure, and tariff policy, the company disclosed this week. The campaign is a new frontier in information warfare: instead of spreading propaganda through social media, Beijing’s operatives weaponized the most popular AI chatbot in the world to influence American policymaking from within.
OpenAI published a report detailing how it identified and disrupted a covert influence operation originating from China. The operatives used ChatGPT to generate persuasive arguments against the construction of AI data centers in the United States, to stoke opposition to federal tech policy, and to shape public opinion on tariffs. The goal, according to OpenAI, was to amplify existing American political divisions around technology and trade in ways that serve China’s strategic interests.
The Axios report on the operation added detail: the influence campaign targeted debates over AI infrastructure spending, data center energy consumption, and the economic costs of tariffs on Chinese goods. By generating large volumes of seemingly organic commentary, the operatives aimed to create the impression of broad grassroots opposition to US tech expansion.
This is a departure from earlier influence campaigns. During the 2016 and 2020 election cycles, Chinese and Russian operatives relied on fake social media accounts, bot networks, and paid trolls to spread disinformation. Those methods are now well understood and increasingly detectable. The use of an AI chatbot to generate persuasive, grammatically correct, and contextually appropriate arguments represents an evolution. The content is harder to detect as inauthentic because it reads like a human wrote it.
OpenAI said it detected the operation through pattern analysis: unusual volumes of ChatGPT interactions originating from coordinated sources, all generating content on the same set of policy topics with the same framing. The company said it terminated the accounts involved and shared its findings with US law enforcement and intelligence agencies.
The timing matters. The United States is in the middle of a massive buildup of AI infrastructure. Tech companies are investing hundreds of billions of dollars in data centers, energy grids, and semiconductor fabrication. The Trump administration has imposed tariffs on Chinese goods as part of a broader decoupling strategy. These are precisely the policies that a Chinese influence operation would want to weaken.
What the report does not say is how effective the campaign was. OpenAI can measure how many ChatGPT queries were generated and what topics they targeted, but it cannot measure how many Americans read the resulting content and changed their minds. Influence operations do not need to convert a majority to succeed. They need only to amplify existing doubts, deepen polarization, and make it harder for policymakers to build consensus.
The broader implication is that AI has given state-backed influence operations a force multiplier. The cost of generating persuasive text at scale is now near zero. The barrier to entry for sophisticated information warfare has dropped from “a room full of paid trolls” to “a single operator with a chatbot.” Every country with a large language model can now run the kind of campaign that once required a dedicated propaganda apparatus.
China’s denial was swift and predictable. Beijing dismissed the OpenAI report as baseless and accused the United States of conducting its own information operations against China. That response is itself part of the pattern: accuse the accuser, muddy the waters, and move on.
The real question is not whether China used ChatGPT to influence US policy. The evidence from OpenAI’s own investigation is credible. The real question is how many other campaigns are running undetected, using AI tools that have not yet learned to flag themselves.