OpenAI Safety has reported the disruption of an influence operation dubbed "Trolling Stone", which utilized AI to generate social media content. The operation focused on the arrest of an alleged Russian cult leader in Argentina, with content supporting the leader, criticizing the Argentinian legal system, and targeting Rolling Stone Argentina magazine.
This case study was initially published in OpenAI’s February 2026 report on disrupting malicious uses of AI.
Key Points
- OpenAI banned ChatGPT accounts involved in the "Trolling Stone" operation.
- The operation generated social media comments about the arrest of Konstantin Rudnev, an alleged Russian "cult leader," in Argentina.
- Activity originated from multiple locations, including Pakistan, Armenia, Uruguay, Argentina, and Kazakhstan.
- A Pakistan-based user was linked to a for-hire actor.
- The operation created short articles about general news in Argentina, posted on Facebook Pages registered on October 14-15, 2025, that posed as news outlets.
- Content included translations of articles about the Rudnev case from Russian into Argentinian Spanish, published on Facebook, YouTube, Medium, and Argentinian news websites.
- ChatGPT models were used to generate comments for these articles, mimicking "astroturfing" behavior.
- Some comments specifically targeted Rolling Stone Argentina for its reporting on Rudnev, including referencing a "refutation" article edited by a Russian-speaking user with ChatGPT on October 26 for publication on October 27.
- Prompts sometimes explicitly requested content that would meet Meta's guidelines.
- Coordination within the network involved users asking ChatGPT to help implement or translate operational instructions, including requirements for posting frequency and daily updates.
Context
According to OpenAI Safety, the "Trolling Stone" operation featured multiple clusters of activity across widely dispersed geographical locations. Despite this spread, the groups demonstrated coordinated behavior, generating pro-Rudnev content in Argentinian Spanish for various online platforms. The operation's content primarily focused on the Rudnev case, either praising him, denying wrongdoing, or criticizing the Argentinian legal system and Rolling Stone magazine.
Why It Matters
This report illustrates how AI models can be misused in coordinated influence campaigns, highlighting the need for vigilance from platform providers and users. It demonstrates how AI can facilitate content generation and coordination in attempts to manipulate public perception.
What To Do
- Note the February 2026 publication date of OpenAI's report on malicious AI uses.
- Observe how ChatGPT was used for both content generation and internal coordination within the influence operation.
- Watch for similar patterns of dispersed but coordinated activity across multiple social media platforms.
- Review the examples of content generation and "astroturfing" tactics described in the case study.
Keep Exploring
/atlas/claude-family /atlas/gpt-family /techniques/role-objective /techniques/constraints /techniques/output-schema /techniques/ask-before-inventing /techniques/multishot /techniques/xml-delimiting /techniques/system-user-separation /techniques/ptcf /techniques/multimodal-grounding /studio?pack=foundation