OpenAI has banned a covert influence operation, likely originating from a private Chinese tech company, for using its models to generate social media campaigns targeting American audiences. The campaign, detailed in a June 2026 threat report, focused on stoking public resentment against US data centers and AI infrastructure by blaming them for rising household electricity costs.
The operation wasn’t just spinning up random text. The operators prompted ChatGPT in Simplified Chinese, requesting outputs that posed as Americans from various backgrounds. They then deployed AI-generated short comments and edited images on X, complete with hashtags like #capacityauction and #datacenters, alongside links to legitimate news stories to give the gripes a veneer of credibility. One specific tactic involved creating comic strips based on a regional newspaper’s reporting about power grid capacity auctions, with the generated text explicitly arguing that costs tied to AI demand were being passed to “ordinary households.”
Beyond the energy narrative, the cluster also targeted overseas Chinese audiences and dissidents, including the activist Li Ying, by attempting to generate insults—though OpenAI notes its models refused those specific inflammatory requests. A more novel tactic involved fabricating personas of US-based Chinese immigrants to encourage a YouTuber, a former Chinese police officer, to produce content criticizing US policy failures.
These weren’t amateurs. The banned accounts used ChatGPT to automate their workflow, writing code for managing logins and interactions across platforms, and they produced internal reports detailing how to build “persistent and credible accounts” designed to evade detection. The operation reveals a maturing, commercially-backed ecosystem that is learning how to weaponize generative AI not for simple spam, but for crafting targeted, multi-platform narratives that exploit specific local grievances.
