OpenAI catches China disguising ChatGPT as Americans: steering public opinion on AI data center electricity prices amid Trump’s tariffs and public complaints

OpenAI bans two clusters of ChatGPT accounts that very likely originate from China. They issue commands in simplified Chinese, disguise themselves as Americans, and attempt to turn the real public grievances over data center electricity costs and U.S. tariff policies into favorable public opinion material for Beijing.
(Background summary: Trump forced the seven major tech giants to sign "bear the costs of electricity price hikes"! AI data centers consume 12% of U.S. electricity, fueling voter anger.)
(Additional background: Anthropic angrily accuses Chinese AI models like DeepSeek of secretly learning from Claude, using 24k fake accounts to flood 16 million questions and answers.)

Table of Contents

Toggle

• Free Riding: How Do the Clusters Operate?
• Americans Oppose Data Center Heating
• What Do Failed Actions Reveal

According to OpenAI’s latest threat intelligence report, there are two "very likely originating from China" ChatGPT account clusters that have long used OpenAI’s models as tools for influencing public opinion. Their targets are not military secrets or elections, but the ongoing AI policy debates within the U.S., the electricity consumption of data centers, and Trump-era tech tariffs.

Free Riding: How Do the Clusters Operate?

OpenAI assigned English code names to the two clusters. The first is called "Data Center Carpool", and the second is called "Technology and Tariffs".

According to the report, operators log into ChatGPT from within China via VPN (since OpenAI does not offer services in China), input commands in simplified Chinese, requesting the model to produce social media posts and images in both English and Chinese, then post them impersonating ordinary Americans on X and YouTube.

More notably, operators even upload operational files explicitly detailing "targets, strategies, and how to fake accounts" directly to ChatGPT to accelerate content production.

The core narrative of the Data Center Carpool cluster is: the AI boom is driving up electricity bills for ordinary households. The report states that this cluster’s output includes a cartoon of a businessman with a cigar, holding a money bag with a dollar sign, next to a family staring in shock at their electricity bill. The text references real news such as "grid operator capacity auctions" and "data center electricity demand," sounding like an angry local U.S. commentary.

Note: This cluster is traced back to an unnamed Chinese tech company that has contracts with multiple local governments in China.

The Technology and Tariffs cluster mainly targets U.S. tariff policies, portraying Trump’s trade protectionism as a "tool to dominate technological competition." The report admits that the prompt for this cluster explicitly states that the output "must not mention Chinese leader Xi Jinping, only President Trump." It disguises itself as American speech, avoiding any Beijing stance. The content covers English, Italian, Japanese, and Traditional Chinese, with the Traditional Chinese version aimed at Taiwanese audiences.

The Technology and Tariffs cluster is also linked to a suspected fake social media network that spreads rumors claiming "ChatGPT user data has been leaked," directly targeting OpenAI. The report emphasizes that these allegations are completely false.

Americans Oppose Data Center Heating

The operator’s focus on the issue of data center electricity costs is not random. Polls show about 71% of Americans oppose data center expansion near their homes; Bloomberg reports that electricity prices near data centers have risen approximately 267% over the past five years. This public resentment does not require external manipulation; it has been there all along.

Currently, AI data centers are estimated to consume about 12% of U.S. electricity. The Trump administration recently pressured the seven major tech giants to commit to bearing the costs of rising electricity prices driven by AI data centers. The operators’ goal is simply to stoke this fire further and frame it as "U.S. tech capital exploiting ordinary families."

Ben Nimmo, head of OpenAI’s threat intelligence, concludes directly: this is not an influence operation "creating" a debate, but riding on an existing one. In simple terms, the operators found a real social fissure, tried to wedge it open further.

The report states that the narratives of the two clusters closely overlap with known Chinese official propaganda themes. However, OpenAI does not directly blame the Chinese government, only saying it is "very likely originating from China," and names the aforementioned Chinese tech company.

What Do Failed Actions Reveal

OpenAI uses an internal "dissemination scale" to evaluate the actual impact of these two clusters. The Data Center Carpool scores 1 point, and the Technology and Tariffs scores 2 points. The scale defines activity as appearing on one or more platforms but with no evidence of meaningful engagement from the target audience. In simple terms, these posts are mostly unseen and unshared.

OpenAI explicitly states that the significance of this operation is not that it changed public opinion, but that it shows Chinese-origin influence operators are testing narratives by turning the U.S. AI infrastructure controversy into a battlefield. However, not breaking through echo chambers does not mean failure; this operational logic is real, being tested and iterated. (It’s hard to say whether OpenAI has identified all related accounts.)

In the final part of the report, OpenAI reaffirms its strategic stance: to build "democratic AI" and oppose the use of AI for surveillance, censorship, and political control—what it calls "AI-featured authoritarianism." This statement is both a framework declaration and a market positioning. It makes this threat intelligence report more than just a technical briefing; it carries a clear geopolitical stance.