OpenAI Accuses China of Manipulating the Data Center Debate

OpenAI recently took action against two groups of ChatGPT accounts, suspected to be based in China, for attempting to manipulate public discussions in the U.S. regarding data centers and tariffs. These groups operated between late 2025 and early 2026, yet failed to generate any measurable authentic engagement. This is the first time OpenAI has traced such an operation back to its own tools in the debate over American data centers.

The investigation led by Ben Nimmo, chief investigator at OpenAI, revealed that these groups used VPNs to bypass access restrictions to ChatGPT, which is banned in China. The first group, dubbed the “Data Center Bandwagon,” produced content in English, including comments and political cartoons, to influence public opinion on the energy costs of data centers. Among the disseminated content were cartoons depicting white-collar workers in suits brandishing bags of money, while families discovered a huge electricity bill due to the expansion of AI infrastructure. This content was shared on platforms like X.com and Facebook.

OpenAI managed to trace back to an unnamed Chinese tech company contracted with several provincial governments. The instructions given to ChatGPT were written in simplified Chinese, but the outputs were requested in both English and Chinese. The operators posed as ordinary Americans on social media.

The second group, “Tech and Tariffs,” targeted Donald Trump's tariff policies, particularly following the announcement of a 100% tax on Chinese goods in October 2025. This group also spread a false allegation regarding a data breach of ChatGPT users, which was denied by OpenAI.

Although OpenAI did not establish a direct link to the Chinese government, the tactics employed by these groups resemble documented Chinese propaganda campaigns. OpenAI's report associates American opposition to data centers with foreign interference, while emphasizing that measured engagement was nonexistent outside the networks of the implicated groups.

Since 2024, OpenAI has published several reports on similar operations conducted by Chinese, Russian, Iranian, and Israeli actors, none of which managed to reach an audience beyond their own network.