OpenAI has banned accounts linked to China that exploited its ChatGPT tool to seed social media with influence campaigns targeting two of America's hottest political debates: tariffs and AI data centers. The company disclosed the discovery Wednesday, revealing what it sees as a harbinger of how state-backed actors plan to weaponize generative AI.
Two distinct operations used ChatGPT to manufacture comments, cartoons, and posts designed to inflame existing tensions. The first, code-named "Data Center Bandwagon," flooded social media with content claiming that AI server farms were spiking electricity costs for ordinary Americans. The second, "Tech and Tariffs," generated political cartoons and commentary attacking Trump-era tariff policies and criticizing American dominance in global technology markets.
Neither campaign gained meaningful traction online, according to OpenAI, but the discovery signals a shift in how foreign powers may attempt to manipulate American public opinion. These weren't efforts to manufacture new grievances from scratch. Instead, the operatives identified fractures that already existed in the American electorate and poured fuel on them using AI-generated content at scale.
A Harvard and MIT survey found 32 percent of Americans oppose data centers in their region, while 40 percent back them. On tariffs, the numbers are starker: a Harris poll from March showed seven in 10 Americans blame Trump's trade policies for higher prices they pay at the store. The Chinese operatives were betting that amplifying these real concerns could deepen political rifts.
"This was not a case of an influence operation creating a debate," said Ben Nimmo, principal investigator on OpenAI's intelligence and investigations team. "The debate existed already. This was an influence operation from China trying to interfere in it."
In the data center campaign, users OpenAI traced to a Chinese government contractor asked ChatGPT to design comic strips about power grids and electricity costs. Those images were posted to X, Twitter's new name, through what appeared to be fake accounts, with links to real news stories about data center power consumption. The approach blended AI-generated fiction with legitimate reporting, a technique designed to bypass automated detection systems.
The second group, which OpenAI could not definitively connect to state actors, generated political cartoons of President Trump wielding a "Tech Dominance" mallet against a wall labeled "Global Future." The imagery attacked both Trump's nationalistic tech agenda and American efforts to maintain technological leadership globally, messaging aimed at audiences skeptical of American exceptionalism.
This marks the first time OpenAI has documented China-linked actors using its models to meddle in debates over AI infrastructure, raising questions about how quickly foreign governments are learning to exploit the latest generation of AI tools. Unlike traditional disinformation campaigns that required armies of fake accounts and human copywriters, these operations demonstrated how a single operator with access to ChatGPT could manufacture dozens of pieces of content in minutes.
The campaigns underscore a broader vulnerability: the United States remains deeply divided on key economic and tech policy questions, and those fault lines are now targets for foreign interference powered by the same AI systems that American companies are racing to dominate.
Author James Rodriguez: "These operations failed to gain real momentum, but that's almost beside the point. What matters is that China's learning the playbook, and they've got time to perfect it."
Comments