OpenAI bans China-linked ChatGPT accounts that amplified US data center electricity price backlash

At a glance:

• OpenAI removed two clusters of China-linked ChatGPT accounts running covert influence campaigns.
• The "Data Center Bandwagon" group created comics blaming AI data centers for rising US electricity bills.
• The "Tech and Tariffs" cluster spread anti-tariff content and fabricated claims about ChatGPT data breaches.

OpenAI has taken down a pair of coordinated influence operations originating from China that used its chatbot models to shape debates around artificial intelligence, energy costs, and US-China trade tensions. The first cluster, dubbed "Data Center Bandwagon," leveraged ChatGPT to produce social media commentary and comic strips that amplified concerns about AI data centers driving up household electricity prices. The second group, called "Tech and Tariffs," generated anti-tariff cartoons and made false claims that user data had been stolen from OpenAI's systems.

Both operations attempted to pose as American voices on the social platform X, using virtual private networks to access ChatGPT in Simplified Chinese while presenting themselves as citizens from diverse backgrounds. OpenAI's investigation found that the Data Center Bandwagon campaign produced virtually no authentic engagement, suggesting the content largely circulated within echo chambers rather than influencing genuine public discourse. The Tech and Tariffs cluster showed slightly more sophistication, generating comment batches in English, Italian, Japanese, and Traditional Chinese targeted at audiences in Taiwan.

The Data Center Bandwagon campaign directly tied itself to real-world energy market dynamics, requesting comic strips about regional electricity capacity auction prices based on reporting from local newspapers. These posts appeared on X alongside hashtags such as #capacityauction and links to legitimate news coverage, creating an illusion of grassroots concern. While no specific grid operator was named in the fabricated content, the underlying anxiety reflects documented power market pressures: PJM Interconnection, which manages the largest US electric grid region, reported that its independent market monitor found an "irreversible" 75.5% increase in wholesale power costs linked to data center demand, with prices near certain data center clusters climbing as much as 267% over five years.

Three US senators have pressed Amazon, Google, and Meta for answers about how rising data center energy costs are being passed through to residential customers, highlighting the political sensitivity of the issue. Ben Nimmo, principal investigator at OpenAI, emphasized that this was not a case where influence operations created a debate from nothing, but rather attempted to amplify existing tensions. OpenAI rated the activity Category One on its Breakout Scale, indicating the campaigns remained confined to a single platform with no evidence of reaching genuine audiences.

The Tech and Tariffs cluster focused on US-China competition themes including tariffs, rare earth minerals, 5G technology, and industrial resilience. Notably, their anti-tariff cartoons depicted former US President Donald Trump but completely avoided showing Chinese leader Xi Jinping, a detail that aligns with previous Chinese influence tactics observed in other campaigns. One operator explicitly referred to their network as a "water army," the Chinese term for coordinated troll networks, and even requested that ChatGPT help design a system for scraping and analyzing social media posts from individuals flagged as potential risks.

OpenAI's models refused to assist with the data collection request, instead returning generic advice about secure data storage practices. The same fake accounts repeatedly posted fabricated claims that ChatGPT user data had been compromised, an apparent attempt to damage OpenAI's reputation. This mirrors tactics seen in earlier operations: OpenAI linked both clusters to the 2022 Spamouflage campaign, which researchers at ASPI and Mandiant found targeting rare earth companies after China's 14th Five-Year Plan prioritized critical minerals. The new activity coincides with the adoption of China's 15th Five-Year Plan, which elevated AI development as a strategic industry.

While these specific influence operations were relatively unsophisticated and generated minimal authentic engagement, they reveal evolving tactics in state-linked information warfare. By using AI models to rapidly produce content in multiple languages and formats, operators can scale their reach while maintaining plausible deniability. The campaigns also demonstrate how geopolitical tensions around emerging technologies can be weaponized, even when the underlying technical claims are exaggerated or false.

Going forward, OpenAI and other AI platform providers face increasing pressure to detect and disrupt such operations before they gain traction. The incident underscores the need for transparency around how AI-generated content is labeled and how platforms verify the authenticity of user interactions across their services.