OpenAI Exposes China-Linked Accounts Stoking US Data Center Fears

There is a particular irony at the center of Engadget's coverage of OpenAI's latest threat report: the influence campaign allegedly run through ChatGPT focused on electricity costs that are, by OpenAI's own admission, genuinely rising. According to a Bloomberg report cited in OpenAI's findings, electricity costs in areas near data centers have climbed as much as 267 percent higher compared to five years ago. Fake voices amplifying a real problem is a harder thing to dismiss than pure fabrication.

OpenAI's report identified two clusters of accounts it believes were operated by a social media team at a private Chinese company working for local government clients. The first, which OpenAI labeled the "Data Center Bandwagon" group, reportedly asked ChatGPT to generate English-language talking points and images, including comic strips, focusing on how AI data centers increase electricity demand and drive up consumer bills. From late 2025 to early 2026, the operators used VPNs, prompted ChatGPT in Simplified Chinese, and posed on X as Americans from various backgrounds. They reportedly uploaded files to ChatGPT detailing their objectives and strategies for creating fake social media accounts without being detected. A second cluster focused on US tariffs and tech policy, generating content in English, Italian, Japanese, and traditional Chinese, with messages critical of the US and its relationships with allies.

The campaigns did not appear to move opinion. OpenAI rated the activity Category One on its Breakout Scale, indicating it stayed on a single platform with no evidence of reaching genuine audiences, and OpenAI admits the campaigns failed to gain much authentic engagement.

What the reporting does not give you is detail on OpenAI's specific attribution methodology for linking the accounts to China, or whether the platforms involved were cooperative in detection and removal. OpenAI is also the organization reporting on misuse of its own systems, which is a transparency worth crediting but comes with the limits of self-reporting.

The more durable finding is the choice of target. Legitimate community opposition to data center expansion, around energy costs and grid strain, exists across the United States. An operation that layers fake amplification onto a genuine grievance is structurally harder to counter than one manufacturing a concern from scratch. For researchers and policymakers tracking how AI tools change the economics of influence operations, that asymmetry is probably the detail worth holding onto.