TL;DR · 30-second read

The Short Version

OpenAI, the company behind ChatGPT, says it shut down a group of accounts, likely run from China, that used its chatbot to write fake American social media posts.

The posts claimed that data centers, the giant computer warehouses that power artificial intelligence, are pushing up ordinary families’ electricity bills.

Worries about power bills are real. But this case shows that the fight over these buildings now draws foreign propaganda, not just debate at local town halls.

OpenAI said on June 1, 2026 that it had banned a cluster of ChatGPT accounts that likely originated in China and used the chatbot to produce content for a covert influence operation it calls “Data Center Bandwagon.” The accounts posed as Americans from a range of backgrounds and posted English-language comments and images on X arguing that data centers and AI applications are driving up electricity demand and passing higher costs to ordinary households. The case study is part of OpenAI’s June 2026 report on PRC-linked influence operations targeting AI debates in the US.

OpenAI assessed that the operators were likely a social media operations team at a private Chinese technology company working for provincial-level government clients. The same cluster also targeted overseas Chinese audiences, including attempts to generate insulting comments about dissident Li Ying. OpenAI says its models refused those requests.

Executive Summary

The takedown is small in scale but specific in subject. According to OpenAI, the operators asked ChatGPT for comic strips about a power grid operator’s capacity auction prices, based on reporting from a legitimate regional newspaper. They wanted the rising prices framed as the result of new peak demand from data centers and AI, with the costs ultimately landing on households. The content was then posted by likely inauthentic accounts on X, next to links to real news stories.

This matters for the infrastructure industry because it moves the data center power-cost debate out of the purely local arena. Until now, the dispute over who pays for the grid upgrades behind AI campuses has mostly played out in utility commission dockets, county zoning hearings and regional press. OpenAI’s case shows that at least one foreign-directed operation treated that same argument as useful material. It sat on the operators’ task list alongside harassing Chinese dissidents.

The case does not show that concern about data center electricity costs is manufactured. OpenAI describes an operation that amplified legitimate reporting while hiding who was doing the talking. It does not describe one that invented the underlying issue. That distinction should shape how developers, utilities and community groups respond.

How the Operation Worked

OpenAI’s account describes a workflow rather than a one-off stunt. The operators prompted ChatGPT in Simplified Chinese and requested outputs in English and Chinese. They reached the service through VPNs, because OpenAI does not allow access from China. Their outputs included short comments, comic strips and edited stock images. For the edits, they added text to generic electricity-market pictures to support a narrative about ordinary people subsidizing AI infrastructure. Posts carried hashtags such as #capacityauction, #datacentersuccess and #datacenters.

The team also used the model as back-office tooling. It asked for code to automate logins and to manage interactions across multiple platforms. It used ChatGPT to extract usernames, prepend X or YouTube links, strip hyperlinks and format data for spreadsheets. It also had ChatGPT polish internal work reports. According to OpenAI, those reports set out objectives such as building persistent, credible accounts and anticipating platform enforcement. In practice, the AI was not doing the persuading. It was lowering the labor cost of running many fake voices at once.

Why Power Bills Were the Chosen Wedge

The operators did not pick an abstract talking point. Capacity auctions are how some regional grid operators pay power plants to be available during the highest-demand hours. Clearing prices in those auctions depend heavily on forecasts of peak load. When forecasts rise, partly because of large new data center loads, the cost of that guaranteed capacity rises too. Those costs generally flow through to retail electricity bills. That is a real, documented mechanism, and it is exactly the chain of cause and effect the operators asked ChatGPT to dramatize: data centers raise peak demand, peak demand raises capacity prices, and households pay.

This is the core of the operational point. The argument over who pays for AI’s grid footprint was already one of the most sensitive issues in data center siting. It turns on cost allocation between large new loads and existing ratepayers, which is complex and easy to reduce to a slogan. That made it a ready-made wedge. The same cluster that pushed this narrative to US audiences was also, per OpenAI, aiming content at overseas Chinese audiences in line with Party-state priorities. So the power-cost debate was treated as one line item in a broader influence effort, not as a niche energy topic.

Everyone who builds, powers or permits AI infrastructure is affected: developers seeking local approvals, utilities filing large-load tariffs, grid operators publishing auction results, and state regulators weighing who should bear upgrade costs. Each now operates in a debate that can be deliberately amplified, and each will find it harder to tell organic sentiment from manufactured sentiment by counting social media volume alone.

Real Concern, Borrowed Voice

It would be a misreading to treat this case as evidence that opposition to data centers is foreign-driven. OpenAI’s own description has the operators linking to legitimate news coverage of capacity auctions and data center demand. The deception lay in who was speaking, fabricated Americans, not in whether rising capacity prices exist. Residents, consumer advocates and local officials raising cost questions are part of a genuine policy debate and deserve answers on the merits. Labeling them as a foreign campaign would be both unfair and unsupported by anything in this report.

The same even-handedness applies to the report itself. OpenAI is one of the largest sponsors of new AI data center capacity, so it has a stake in how this debate goes. Its attribution rests on behavioral evidence: prompts in Simplified Chinese, VPN access, uploaded work reports describing detection evasion, and coordinated persona use. That evidence speaks to coordination and inauthenticity, not to whether the cost argument is right. OpenAI also does not say how far the content spread. The case shows intent and method. It does not show impact.

What Changes for Builders and Platforms

For the infrastructure industry, the practical lesson is that vague reassurance is a weak defense in a debate that can be deliberately amplified. Projects that publish verifiable figures give neutral observers something to check a viral claim against: contracted load, tariff structure, who funds transmission and substation upgrades, and how capacity costs are allocated. The more concrete the public record, the less room a fabricated narrative has to work in.

For platforms, the most revealing material is the planning document. OpenAI says one report focused on Facebook and laid out a plan to build lifestyle and commentary personas, amplify narratives through cross-account interaction while preserving the look of organic engagement, and combine organic posting with paid ads. It also called for backup accounts and separated account activity to avoid detection. That is a playbook for long-term presence, not a burst of spam, and it suggests detection will increasingly depend on cross-company signal sharing rather than on any one platform’s view.

Background

OpenAI publishes periodic threat reports describing accounts it has banned for using its models in influence operations, scams and cyber activity. The Data Center Bandwagon case is one entry in its June 2026 report on PRC-linked operations targeting US AI debates, and it follows an earlier report in which OpenAI documented harassment of the same dissident, Li Ying.

The target topic is one of the most contested in AI infrastructure. Large AI data centers can draw as much electricity as a sizable town. Their growth has fed into higher peak-demand forecasts, and in some regions into higher capacity auction prices, which in turn reach household bills. Utilities, grid operators, regulators and communities have been debating how those costs should be shared between new large loads and existing customers.

Sources

Source: “Data Center Bandwagon” Campaign: US-targeted influence activity – OpenAI, OpenAI’s case study on a banned ChatGPT account cluster that generated social media content about data center electricity costs.