OpenAI Bans China-Linked Accounts Using ChatGPT for U.S. Influence

OpenAI has shut down China-linked accounts that weaponized ChatGPT to manufacture social media content designed to inflame U.S. political debates over tariffs and artificial intelligence infrastructure, the company disclosed Wednesday.

The operations generated posts, comments, and political cartoons aimed at exploiting existing tensions in American tech policy discussions. While the campaigns failed to achieve meaningful engagement, they represent an emerging pattern of foreign actors testing generative AI tools to scale influence operations.

Two distinct operations

OpenAI identified two separate campaigns that leveraged its language model to produce content at scale. The first operation, which the company labeled "Data Center Bandwagon," focused on AI infrastructure concerns. Users believed to be connected to a Chinese government contractor prompted ChatGPT to create comic strips suggesting that AI data centers were driving up electricity costs for American households. These images were subsequently posted to X through accounts OpenAI assessed as likely inauthentic, accompanied by links to legitimate news coverage of data center power consumption.

The second campaign, termed "Tech and Tariffs," generated content criticizing former President Trump's tariff policies and broader U.S. technology strategy. One political cartoon depicted Trump in American flag pants labeled "America First," wielding a mallet marked "Tech Dominance" against a wall reading "Global Future." OpenAI could not directly attribute this second operation to a specific entity.

Why it matters

These campaigns signal how adversarial nations may exploit generative AI to amplify divisions around genuine policy debates. The operations didn't create controversies from nothing—they parasitically attached to real concerns. Recent polling shows Americans are already divided on these issues: 32% oppose data centers in their communities while 40% support them, according to Harvard and MIT researchers. Separately, 70% of Americans told Harris pollsters in March that Trump's tariffs had increased their costs.

Ben Nimmo, principal investigator on OpenAI's intelligence and investigations team, emphasized this distinction: "This was not a case of an influence operation creating a debate. The debate existed already. This was an influence operation from China trying to interfere in it."

The campaigns' failure to gain traction offers limited comfort. OpenAI officials noted this appears to be the first instance they've observed of China-linked actors using their models to target the AI data center debate specifically—suggesting adversaries are actively probing which American policy flashpoints are vulnerable to AI-augmented influence operations.

The scaling problem

The core threat isn't sophistication but scale. Generative AI tools can produce vast quantities of contextually relevant content far faster than human operators, lowering the barrier for sustained influence campaigns. Even operations that fail individually contribute to a broader strategy of testing which narratives and tactics resonate with American audiences.

OpenAI says these incidents represent early indicators of how foreign influence operators may deploy AI tools against U.S. political pressure points in the future. The company has banned the associated accounts.

These details were first reported by Axios.