OpenAI disclosed a coordinated effort by a group of accounts it believes were linked to a private Chinese firm working for local‑government clients. The actors used the chatbot to draft English‑language talking points, images and even comic strips that framed AI data centers as a driver of higher electricity costs for American consumers. By posing as U.S. residents from diverse backgrounds, they posted the generated material on social media platforms, hoping to insert themselves into an ongoing discussion about the nation’s AI infrastructure.
The report divides the operation into two clusters. The first, labeled “Data Center Bandwagon,” focused on highlighting the strain that AI‑driven data centers place on regional power grids. Users asked ChatGPT to produce narratives about how these facilities push up demand, leading to higher utility bills. The output included visual assets such as infographics and comic strips that were then shared alongside links to legitimate news stories about power‑grid capacity auctions and data‑center energy consumption.
Beyond the U.S. audience, the campaign also targeted Chinese expatriates and critics of the Chinese government. In some cases, the actors instructed the model to generate insults aimed at dissidents and political commentators. They masqueraded as Chinese immigrants and professionals living in America, urging public figures to comment on perceived U.S. policy failures.
The second cluster of accounts produced content critical of U.S. tariffs and technology policies, portraying the United States as untrustworthy toward its allies. This group instructed ChatGPT to avoid references to Chinese President Xi Jinping in the images they generated and to write comments in multiple languages—including English, Italian, Japanese and traditional Chinese—to reach Taiwanese audiences.
OpenAI says the campaigns achieved little authentic engagement and did not significantly alter public sentiment. The report notes that the operators chose topics already contentious in the public sphere, which limited their impact. Nonetheless, OpenAI emphasizes the significance of the attempt: actors tried to covertly embed themselves in a domestic debate about AI capabilities while masking their identities and motives.
While the deceptive activity did not succeed in reshaping opinion, the incident raises concerns about the ease with which generative AI tools can be weaponized for misinformation. OpenAI declined to speculate on why the operators selected its chatbot over alternative models, stating it could not determine the underlying motivation.
Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.