China-Linked Influence Op Got Articles Into Mainstream Latin American Media — OpenAI Disrupts "Sponsored Discontent"

Hi, it's Shiichan! Today I've got a slightly unsettling story from an OpenAI report on countering misuse of AI.

OpenAI News openai.com

What was announced?

OpenAI's News reported that it banned ChatGPT accounts linked to a likely China-origin influence operation. The accounts were used to generate short comments in English and long-form news articles in Spanish. Their activity followed a timeline consistent with mainland Chinese business hours, and the Spanish-language articles ended up published on real news sites in Latin America. OpenAI named the campaign "Sponsored Discontent," after finding that some of the articles carried a "sponsored" (patrocinado) label.

Why it matters

This takedown stands out because, according to OpenAI, it's the first time they've observed a likely China-origin influence actor successfully getting articles placed in mainstream Latin American outlets. Some articles carried that "patrocinado" (sponsored) label, and the actor also had ChatGPT translate what looked like Spanish-language receipts from a local news organization — both signs pointing to a paid placement arrangement. Beyond the usual playbook of flooding social media with posts, this shows a newer route into real media outlets, which matters a lot for how AI misuse gets tracked.

What changes

By banning the accounts, OpenAI cut off this network's ability to keep generating new content with ChatGPT. OpenAI uses a tool called the Breakout Scale (rated 1 to 6) to gauge how far an influence operation actually spread, and it rated this campaign as Category 4 - breakout to mainstream media. That's a notably higher rating than operations confined to social media, and while OpenAI can't independently confirm how many people actually read the articles, it flags that this campaign may have reached a meaningfully wider audience among everyday readers in Latin America.

Dive Deep

The operation's output fell into two main buckets.

• Short comments in English and Chinese: criticism of Cai Xia, a dissident and former professor at the Chinese Communist Party's Central Party School. Some comments questioned her expertise on Taiwan or Chinese history and literature; others alleged visa irregularities. This overlapped in part with "Spamouflage," a separate operation OpenAI reported on back in May. The comments were posted from X (formerly Twitter) accounts posing as people from the US or India
• Long-form articles in Spanish: focused on social and political division within the US, covering inequality, political violence, discrimination and sexism, crime, criticism of US foreign policy, and social crises like drug use and homelessness. The content was broadly bipartisan. The actor's method was to have ChatGPT translate and expand publicly available Chinese-language articles into Spanish

The posting activity was especially heavy around the 2024 APEC Forum held in Peru. Articles ran mainly on Peruvian news sites, but also in Mexico and Ecuador, and almost all of them carried the byline of the same individual. Some publications identified that individual as the legal representative of a Chinese company, Jilin Yousen Culture Communication Co., Ltd. (优森文化有限公司).

On the X comment side, the accounts targeting Cai Xia had low follower counts and showed no sign of significant engagement. The Spanish-language articles were a different story - OpenAI identified 19 published articles matching this actor's output, including headlines like:

• "The Ugly Truth of How the US Benefits From Aiding the War in Ukraine" (La Republica, Peru)
• "The Tragic Reality of America's Homelessness Problem" (Wapa, Peru)
• "The Crazy Logic of US Sanctions Against a Palestinian Activist" (El Popular, Peru)
• "Hurricane 'Helene': False Statements and Conspiracy Theories From US Politicians" (Wapa, Peru)
• "America's Food Security Problem and Its Impact on Schools" (Extra, Ecuador)

One article was also identified on a Spanish outlet, El Plural.

Wrap-up

• OpenAI banned ChatGPT accounts linked to a likely China-origin influence operation dubbed "Sponsored Discontent"
• It generated two types of content: short English/Chinese comments attacking dissident Cai Xia, and long-form Spanish articles on US social and political division
• The Spanish articles ran as 19 pieces across real outlets in Peru, Mexico, Ecuador, and Spain, some labeled "sponsored," suggesting paid placement
• OpenAI rated it Category 4 on the Breakout Scale (breakout to mainstream media) - the first confirmed case of a China-origin operation landing articles in mainstream Latin American media
• The X comment campaign, by contrast, saw little engagement and no real spread

If you're curious about the newer ways AI misuse can slip past the usual defenses, this report is worth a read!