The San Francisco-based company said the accounts were used to produce social media comments, comic strips, slogans and political cartoons for apparent covert influence operations. The activity was assessed as likely originating from China and was removed after internal investigators found the accounts had used OpenAI’s models to support narratives aimed at inflaming domestic disputes in the United States.
The data centre effort, named “Data Center Bandwagon” by OpenAI, focused on complaints that AI facilities were straining power grids, raising utility costs and harming communities. Operators asked ChatGPT to draft English-language posts and images, often while prompting the system in Simplified Chinese and using virtual private networks. The content was then pushed through likely inauthentic accounts posing as Americans.
OpenAI said the operation failed to gain meaningful traction beyond its own activity. That assessment limits the immediate political impact of the campaign, but the case has drawn attention because it shows how foreign operators are experimenting with generative AI to scale narratives around sensitive infrastructure projects. The company said the campaign did not create opposition to data centres; it attempted to exploit concerns that already existed.
A second cluster, called “Tech and Tariffs”, generated material criticising US tariff policy and efforts to dominate technological competition. Some prompts instructed the model to depict President Donald Trump while avoiding references to China’s leader Xi Jinping. The cluster also produced content in multiple languages and was connected to claims that ChatGPT user data had been compromised, an allegation OpenAI rejected as false.
The activity appears to have run from late 2025 into early 2026. One group involved in the data centre campaign was assessed as likely linked to a private technology firm working for provincial-level government clients in China. The operators also sought help with operational tasks, including polishing work reports and discussing ways to avoid detection by social media platforms.
The China embassy in Washington said it was not familiar with the report and rejected what it described as groundless attacks and smears. Beijing has repeatedly said it supports responsible use of artificial intelligence and opposes disinformation, while Washington and technology companies have grown more vocal about covert influence activity tied to strategic competition over AI.
The disclosure comes as data centre construction has become a sharper political issue in the United States. Generative AI systems require vast computing capacity, and the facilities that train and run them need large volumes of electricity, water and land. Global data centre electricity consumption is projected to roughly double by 2030, reaching about 950 terawatt-hours and accounting for around 3 per cent of global demand. AI-focused facilities are expected to grow faster than the wider sector.
US power demand is also rising after years of comparatively flat growth. Electricity consumption is forecast to hit record levels in 2026 and 2027, with commercial use overtaking residential demand for the first time as data centres, manufacturing and electrification lift load on the grid. Local opposition has grown in parts of Virginia, Georgia, Arizona, Texas and other states where communities are questioning water use, noise, land deals, tax incentives and grid costs.
That context makes the OpenAI findings politically sensitive. Supporters of rapid AI infrastructure expansion argue that foreign-linked actors are trying to slow US technological leadership by amplifying environmental and consumer concerns. Critics of the industry counter that foreign manipulation should not be used to dismiss genuine local grievances over power bills, planning transparency and resource consumption.
OpenAI’s findings give limited support to warnings that outside actors are seeking to influence the data centre debate, but they do not show that public opposition is primarily foreign-driven. The company’s own assessment indicates the accounts had little visible effect, reinforcing a broader pattern in which generative AI helps operators produce more content without necessarily producing more persuasion.
Follow Arabian Post
Select Arabian Post as your preferred source on Google and MSN News for trusted business news and Arab politics and updates.
