China's Influence Operators Used ChatGPT to Manufacture the Backlash Against AI Data Centers

The anger at AI data centers was partly manufactured, and written by ChatGPT

In its June 2026 threat report, OpenAI disclosed that it had banned two clusters of ChatGPT accounts, likely operated from China, that used the chatbot to fabricate grassroots-looking outrage in the United States. One network churned out comments and cartoons blaming AI data centers for rising household electricity bills. The other attacked US tariffs and, more pointedly, spread false claims that ChatGPT itself had leaked its users' data.

Neither campaign caught fire. OpenAI rated both at the bottom of its severity scale, with little or no real engagement. The reason this still matters is the target. For the first time in these reports, foreign influence operators were caught road-testing narratives against AI infrastructure itself, treating the fight over who pays for the data center boom as terrain worth capturing.

What OpenAI found

The two networks were distinct but similar in method. OpenAI named the first "Data Center Bandwagon." Its operators, assessed to be a social media team at a private Chinese technology firm working for provincial government clients, generated English-language posts and comic strips arguing that the electricity demand from data centers was being pushed onto ordinary American families. They built the comics around real reporting from a regional US newspaper about a power grid operator's capacity auction (the market where utilities buy the right to supply electricity during peak demand), then posted them on X under hashtags like #datacenters and #capacityauction. The second network, "Tech and Tariffs," produced political cartoons attacking US technology policy and instructed the model to draw only President Trump, never China's leader Xi Jinping.

What makes this worth understanding is the mechanism, because no system was hacked. An influence operation is a content factory, and the operators used ChatGPT as the machinery on the line. Blocked from OpenAI's services inside China, they logged in through VPNs (virtual private networks, which disguise where a user is connecting from) while posing as Americans. They asked the model for large batches of short, colloquial comments, translated their messaging into Italian, Japanese and Traditional Chinese to reach audiences including Taiwan, edited stock images to carry their slogans, and even had ChatGPT write code to automate logging into accounts and posting across platforms. One cluster referred to its own fake profiles by the Chinese term "water army," the label for coordinated accounts that flood a platform to simulate a crowd.

• OpenAI assessed both campaigns as Category One on its Breakout Scale, meaning the activity stayed on a single platform with no evidence of spreading.
• The claims that ChatGPT user data had been compromised were, in OpenAI's words, entirely false, and appear to have been an attempt to damage the company's reputation.
• The model refused some requests, declining to generate personal attacks on a named Chinese dissident and withholding help to design a system for surveilling "key persons" online.

Why a failed campaign should still worry you

The concrete outcome here is almost reassuring: accounts banned, engagement negligible, the breach claims fabricated and debunked. Read it that way and you miss the point. The operators were not really trying to win the data center argument on any particular day. They were probing whether a genuine American anxiety, the rising cost of power, could be hijacked to turn people against the infrastructure their country's AI lead depends on. That technique travels. If your organization builds data centers, runs energy projects, or sits anywhere near a politically charged industry, the same playbook can manufacture a synthetic local opposition to your next site, complete with translated talking points and cartoon-ready outrage. The invented data breach is its own warning: a coordinated network can now fabricate a credible-looking scandal about your company at the speed of a chatbot. The systemic shift is the quiet irony OpenAI itself noted, that operators used America's leading AI to attack America's AI. Detection no longer hinges on spotting broken grammar or a foreign accent in the text, because the text is fluent and native. It hinges on behavior: the burst-created accounts, the VPN origin, the coordinated posting.

How to defend against it

1. Watch for coordinated behavior, not just bad content. Track sudden clusters of new accounts pushing the same framing on an issue that touches your organization, because fluent AI-written posts no longer betray themselves through language and only the pattern of activity gives them away.
2. Pre-write your response to a fabricated scandal. Have a plan to rapidly and publicly refute a false breach or safety claim, since OpenAI's experience shows a made-up allegation can be seeded and amplified before you would otherwise notice.
3. Brief the teams that shape public narratives. Make sure communications, public affairs, and trust and safety staff know that grassroots opposition to a project can be synthetic, so a wave of "local" complaints gets verified before it drives a decision.

Bottom line

The campaigns OpenAI shut down failed on their own terms, but they show where foreign influence work is heading: cheap, multilingual, native-sounding, and aimed squarely at the AI industry and the debates around it. The question to raise in your next meeting is not whether someone could fake a groundswell against your company or your infrastructure. It is whether you would be able to tell the difference between real public concern and a crowd that a chatbot wrote overnight.