OpenAI Bans Accounts Linked to “Trolling Stone” Operation
OpenAI reported on February 1, 2026, that it removed multiple ChatGPT accounts associated with an influence operation named “Trolling Stone.” These accounts were utilized to generate batches of social media posts and short news-style articles regarding the arrest of Konstantin Rudnev, an alleged Russian “cult leader,” in Argentina.
Operation Overview
The operation produced short articles—typically three to four paragraphs—about Argentine news and translated Russian coverage of the Rudnev case into Argentinian Spanish. These texts were posted on Facebook Pages impersonating local news outlets, with several Pages registered on October 14–15, 2025. OpenAI noted that four out of six Pages displayed admin locations in Facebook’s transparency settings, with three managed from Uruguay, Armenia, and Pakistan, and a fourth from Pakistan and Uruguay.
Content Generation and Coordination
OpenAI indicated that the network generated both supportive content for Rudnev and critical posts targeting the Argentinian legal system and media coverage, including responses aimed at Rolling Stone Argentina. Some accounts produced comments in Argentinian Spanish and disseminated them across Facebook, Instagram, YouTube, and Medium. The report highlighted that some published articles closely resembled text generated by ChatGPT, with minor edits such as the removal of em-dashes.
Operational Tactics
In one instance, OpenAI reported that a ChatGPT user edited a “refutation” article the day before its publication, with the editor identified as a Russian-speaking participant in the operation. The report also detailed internal coordination efforts, noting that a user based in Pakistan requested ChatGPT to implement operational instructions, which included a target of 20 posts per fake account daily and daily progress updates. Other prompts sought translations, content tailored to meet Meta’s guidelines, and promotional material referencing esotericism, shamanism, yoga, and meditation—activities described in open-source reporting as recruitment tactics for Rudnev’s alleged “cult.” OpenAI stated it could not independently verify these claims of activity.
Impact Assessment
OpenAI assessed the campaign as being toward the low end of Category 4 on its Influence Operation (IO) Breakout Scale. This assessment was based on the observation that most Pages had only a few hundred followers and typical posts received single-digit engagements, although some articles did appear on regional Argentine news outlets. Following the identification of this coordinated activity, OpenAI banned the associated ChatGPT accounts.
Original source: OpenAI News