PRC-linked influence operations are targeting AI debates in the US

Overview

OpenAI said it banned two China-linked covert influence operations that used ChatGPT to generate social media comments, images, and political cartoons aimed at inflaming US public debates over AI data centers and trade tariffs. Both campaigns scored Category One, the lowest tier on OpenAI's Breakout Scale, meaning they stayed on a single platform and reached no genuine audience. The findings, published June 10, 2026, show how state-linked actors are testing AI tools to manufacture political content cheaply, even when the real-world reach stays small.

Key Takeaways

OpenAI identified and removed two covert clusters tied to China, named Data Center Bandwagon and Tech and Tariffs, that used ChatGPT to target ongoing US policy arguments.
Data Center Bandwagon produced comments and comic strips claiming AI data center expansion was raising household electricity bills for ordinary families.
Tech and Tariffs generated cartoons attacking US tariffs as a bid for technology dominance, with prompts instructing the model to feature President Trump and leave out Chinese leader Xi Jinping.
Both operations rated Category One on OpenAI's Breakout Scale, the lowest level, and showed no sign of reaching real people or shifting US opinion.
Operators reached ChatGPT through virtual private networks from inside China, prompted the system in Chinese, and requested output in both English and Chinese.
OpenAI traced one cluster to a private Chinese technology firm working for provincial-level government clients but said it lacked enough evidence to firmly attribute the cartoon campaign.

Stats & Key Facts

#2 China-linked influence operations were identified and banned in the report.
#Both campaigns scored Category One, the lowest of the tiers on OpenAI's Breakout Scale.
#At least 36 data center projects were blocked or delayed in the US between May 2024 and June 2025.
#Data centers accounted for about 1.5 percent of global electricity use in 2024.
#Data center electricity consumption has grown about 12 percent per year over the past five years.

Data Center Bandwagon Targeted Anger Over Electricity Bills

The first cluster zeroed in on a live US worry: rising power costs tied to large data center builds.

›Operators asked ChatGPT to produce comic strips showing households shocked by their power costs.
›The content blamed AI data center expansion for higher electricity bills hitting ordinary families.
›OpenAI linked this cluster to people likely working at a private Chinese technology company serving provincial government clients.
›Images and comments were posted to X through likely inauthentic accounts using tags such as #capacityauction and #datacenters.

Tech and Tariffs Attacked US Trade Policy in Cartoons

The second cluster focused on framing US tariffs as an aggressive play for technology power.

›The campaign generated comments and political cartoons criticizing US tariffs as an attempt to dominate global technology competition.
›Prompts told the model to feature only President Trump and to leave Chinese leader Xi Jinping out of the images.
›That instruction points to deliberate political framing built for US audiences rather than Chinese ones.
›OpenAI said it did not have enough evidence to firmly attribute this cartoon campaign to a specific group.

Why Both Campaigns Scored Category One and Failed to Spread

OpenAI grades the reach of covert networks on its Breakout Scale, and these two ranked at the bottom.

Category One is the lowest tier on the scale. It means the activity stayed on a single platform and showed no sign of breaking out to a genuine audience. OpenAI said it found no evidence either effort meaningfully influenced US opinion.

Ben Nimmo, the principal investigator on OpenAI's intelligence and investigations team, said this was not a case of an influence operation creating a debate. Instead, the operators latched onto arguments Americans were already having and tried to steer them. The activity appeared to have little or no real-world effect.

How Operators Reached ChatGPT From Inside China

OpenAI does not permit access to its models inside China, so the operators worked around that block.

›Operators used virtual private networks to reach ChatGPT from within China.
›They prompted the system in Chinese, including simplified Chinese, while requesting output in English and Chinese.
›One operator self-identified their accounts as a water army, a term for coordinated paid troll networks common in China.
›OpenAI traced the behavior across its systems and banned the accounts involved.

The Real US Tension Over Data Center Growth

The campaigns exploited friction that already exists around AI infrastructure in the United States.

›Researchers counted at least 36 US data center projects blocked or delayed between May 2024 and June 2025.
›Data centers made up roughly 1.5 percent of global electricity use in 2024.
›Their power consumption has grown about 12 percent per year over the past five years.
›Local pushback over energy prices and grid strain gave the operators a ready-made grievance to amplify.

What This Means for Business Readers Watching AI Risk

The episode is less about reach and more about how cheaply state-linked actors are testing AI for influence work.

The takeaway is not panic over impact, since both campaigns flopped on reach. The signal is cost and scale. Generative AI lets a small team produce comments, edited images, and cartoons quickly, then post them through inauthentic accounts to test what sticks.

For business owners, the practical lesson is to treat viral grievance content, especially around hot topics like energy costs or trade, with healthy skepticism. The same tools that draft a marketing email can draft a propaganda comic. OpenAI's disclosure is part of an ongoing threat-intelligence program that removes and documents these networks as it finds them.

Frequently Asked Questions

What did the two China-linked campaigns try to do?

They used ChatGPT to create social media comments, images, and political cartoons that inflamed existing US debates. One pushed the idea that AI data centers raise electricity bills, and the other attacked US tariffs as a grab for technology dominance.

Did the influence operations actually work?

No. OpenAI rated both Category One on its Breakout Scale, the lowest tier, meaning they stayed on one platform and reached no genuine audience. The company found no evidence they meaningfully shifted US opinion.

How did operators in China reach ChatGPT if OpenAI blocks access there?

They used virtual private networks to mask their location and reach the service from inside China. They prompted the model in Chinese while asking for output in both English and Chinese.

Who was behind the campaigns?

OpenAI linked one cluster, Data Center Bandwagon, to a private Chinese technology firm serving provincial-level government clients. It said it lacked enough evidence to firmly attribute the second cartoon campaign, Tech and Tariffs.

Why does this report matter if the campaigns failed?

It shows how cheaply and quickly state-linked actors can now produce political content with AI tools, even when the reach stays small. The concern is the scale and low cost of testing influence tactics, not the impact of these specific efforts.

OpenAI's disclosure shows state-linked actors testing AI to mass-produce political content aimed at US audiences, even though both campaigns stayed small and failed to gain traction. The bigger story is how cheap and fast such influence attempts have become.