OpenAI disrupts first Chinese influence op that planted AI-written articles in Latin American media

OpenAI has banned a cluster of ChatGPT accounts linked to a covert influence operation it’s calling “Sponsored Discontent,” marking the first time the company has observed a likely China-origin actor successfully planting long-form, AI-generated articles in mainstream Latin American news outlets. The operation used OpenAI’s models to produce Spanish-language articles critical of the United States, which then appeared in publications across Peru, Mexico, and Ecuador, often on the same day they were generated.

The accounts followed a workflow timed to mainland Chinese business hours and produced two distinct content streams. One involved short, English-language social media posts attacking Chinese dissident Cai Xia, a pattern consistent with the previously identified “Spamouflage” operation. The other, more novel effort focused on translating and expanding publicly available Chinese articles into long-form Spanish pieces that highlighted American social crises—from political violence and inequality to drug use and homelessness—framing them as evidence of failed U.S. leadership. The campaign was broadly bipartisan, criticizing both major American parties.

What distinguishes this operation is not just the content, but the access it bought. OpenAI’s open-source research found that at least one of the articles was listed as “sponsored,” and the actor used ChatGPT to translate financial receipts from a Latin American news organization, strongly suggesting a pay-to-play arrangement. Almost all the articles were attributed to the same individual, whom some outlets identified as the legal representative of a Chinese company called Jilin Yousen Culture Communication Co., Ltd. This is a significant escalation from standard bot-driven social media spam; it moves the fight for narrative control from the chaotic feed of X to the relatively trusted real estate of local news websites.

While the social media posts targeting Cai Xia garnered little authentic engagement—accounts typically had low follower counts and minimal interaction—the impact of the Spanish-language articles is harder to dismiss. OpenAI applied the Breakout Scale, a framework for rating influence operations, and assessed this campaign as a Category 4: “breakout to mainstream media.” This rating reflects the wide readership of the regional news sites, even if precise engagement figures are impossible for OpenAI to independently verify. The operation was particularly active in the lead-up to the 2024 APEC Forum in Peru, suggesting a deliberate, opportunistic attempt to shape regional perceptions of the U.S. during a high-profile diplomatic event.