OpenAI Exposes Covert AI Influence Networks
According to an investigation report from OpenAI cited by The Decoder, the company exposed and banned ChatGPT accounts involved in coordinated Russian and Iranian influence operations. Rather than relying on traditional social media campaigns, these threat actors used fake identities to plant fabricated content directly into legitimate news organizations.
The Russian operation, identified as Dark Clark, disseminated disinformation across Latin America intended to discredit Ukraine and destabilize local politics. Operating through a fictitious persona controlling a think tank, the actors generated fabricated audio files and documents that successfully triggered official fact-checks and political reactions in Ecuador and Peru. OpenAI rated this campaign a category 5 on the Breakout Scale due to tangible political reactions, marking the first time in two and a half years of reporting that a network reached this severity level.
Concurrently, the Iranian operation named Bogus Bylines deployed seven fake journalists to place nearly one hundred articles concerning the United States and Iran conflict across online outlets worldwide. While both networks utilized large language models primarily for internal reporting and adapting propaganda into multiple languages, the operations highlight the evolving integration of generative technology in geopolitical influence efforts.
Based on reporting by the-decoder.com.
