OpenAI Disrupts 'Sponsored Discontent' Influence Operation
OpenAI has identified and banned ChatGPT accounts associated with a Chinese influence operation named "Sponsored Discontent." This operation used AI-generated content to target a Chinese dissident and plant critical narratives about the United States in mainstream Latin American news outlets, marking a significant shift in the tactics of China-origin influence operations.
Operation "Sponsored Discontent" Overview
OpenAI discovered a campaign that utilized ChatGPT to generate two distinct types of content: short-form English comments targeting a Chinese dissident and long-form Spanish articles critical of the US. The activity followed a timeline consistent with mainland Chinese business hours and the accounts prompted ChatGPT in Chinese.
Actor Attribution and Tactics
The activity is linked to a likely Chinese actor, with some articles attributing content to an individual connected to the Chinese company Jilin Yousen Culture Communication Co, Ltd (优森文化有限公司).
Key tactical observations include:
- Content Generation: The actor used ChatGPT to translate and expand existing Chinese-language articles into Spanish.
- Distribution: Long-form content was published by news sites in Latin America, specifically in Peru, Mexico, and Ecuador.
- Financial Arrangements: Evidence suggests the use of paid publications, with some articles explicitly listed as "sponsored" (patrocinado) and the actor using ChatGPT to translate Spanish-language receipts from news organizations.
Content Strategies and Targets
Targeting of Cai Xia
In parallel with the broader US-critical campaign, the actor generated short criticisms of Cai Xia, a former professor at the Chinese Communist Party’s Central Party School and dissident. The operator prompted ChatGPT to respond to tweets or videos posted by Cai Xia, or media mentions of her, challenging her expertise in Chinese literature, history, Taiwan, and the internet, or accusing her of visa irregularities. This activity is consistent with the covert influence operation known as "Spamouflage."
US-Critical Narratives in Latin America
The long-form Spanish articles focused on social and political divisions within the United States, often timed to coincide with events like the 2024 APEC Forum in Peru. The content portrayed US social crises‣such as drug use, homelessness, inequality, and political violence‣as evidence of a failed society or failed leadership. The narratives were broadly bipartisan, criticizing both major US political parties.
Impact Assessment
Social Media Engagement
OpenAI found no evidence that the content targeting Cai Xia on social media achieved significant authentic engagement, as the posting accounts typically had low follower counts and minimal interaction.
Mainstream Media Penetration
Unlike typical social media campaigns, "Sponsored Discontent" successfully planted articles in mainstream Latin American media. OpenAI assesses this as a Category 4 on the Breakout Scale (breakout to mainstream media) due to the wide readership of these news websites. This represents a previously unreported line of effort for China-origin influence operations, potentially reaching a significantly wider audience than traditional social media activity.
Identified Published Articles
OpenAI identified several articles matching content generated by this actor, including pieces published in La Republica (Peru), Wapa (Peru), El Popular (Peru), Libero (Peru), El Universal (Mexico), and Expreso (Ecuador). Topics ranged from the US role in the Ukraine war and the fentanyl crisis to the security of children and the failure of US democracy.