Operation “Sponsored Discontent”: Influence activity

This case study was originally published in OpenAI’s February 2025⁠(opens in a new window) report.

Actor

We banned ChatGPT accounts that used our models to generate short comments in English and long-form news articles in Spanish. This activity followed a timeline consistent with mainland Chinese business hours. The Spanish-language articles were published by news sites in Latin America. Almost all the websites attributed the articles to an individual; in some cases, they connected that individual to a Chinese company, Jilin Yousen Culture Communication Co, Ltd (优森文化有限公司), in the byline of the articles.

Behavior

The accounts prompted ChatGPT in Chinese and were mostly active during mainland Chinese business hours. Some activity was consistent with the covert influence operation known as “Spamouflage⁠,” which we reported on in May: it generated short English-language comments and images that criticized Cai Xia, a dissident and former professor at the Chinese Communist Party’s Central Party School. This content was then posted on X by accounts which posed as people from the United States or India.

If this had been the only activity, we might have been led to conclude that it was likely linked to “Spamouflage.” However, alongside these short comments, the accounts also generated long-form articles in Spanish that were critical of the US. This activity was particularly prevalent in the lead-up to the 2024 APEC Forum in Peru. The actor generated these articles by asking our models to translate and expand publicly available Chinese-language articles.

These articles were then published by Latin American news websites, especially in Peru, but also Mexico and Ecuador. The news websites often published the articles on the same day the accounts generated the content using ChatGPT. Almost all the publications attributed the articles to the same individual. Some identified the individual as the legal representative of what we believe to be a Chinese company.