OpenAI has taken action against several influence operations, including the Russia-origin operation "Doppelganger," by banning associated accounts that used its AI models to generate anti-Ukraine social media comments, translations, and website copy across multiple languages. This case study was initially published in OpenAI's May 2024 report. In subsequent reports, OpenAI also detailed disruptions of "A2Z" in October 2024 and "Sponsored Discontent" in February 2025.
The "Doppelganger" operation targeted audiences in Europe and North America, focusing on content generation for websites and social media. The accounts displayed varied tactics, techniques, and procedures, suggesting an operation composed of different functional teams.
Key Points
- OpenAI banned four clusters of accounts linked to the Russia-origin "Doppelganger" operation.
- The first cluster generated short text comments in English, French, German, Italian, and Polish, posted on 9GAG and X.
- This cluster accessed OpenAI's models via a service linked to companies in Russia and the Czech Republic, which OpenAI has since disabled.
- The second "Doppelganger" cluster translated Russian articles into English and French, publishing them on rrn.media.
- The third "Doppelganger" cluster generated, proofread, and corrected French-language articles for franceeteu.today.
- The fourth "Doppelganger" cluster operated in German, generating posts for a Facebook Group and Page from grenzezank.com content.
- The "A2Z" operation, detailed in OpenAI's October 2024 report, used OpenAI's API to create comments for X and Facebook, often praising Azerbaijan.
- The "Stop News" operation, also from Russia and detailed in OpenAI's October 2024 report, used ChatGPT to generate short comments, long-form articles, and images in English, French, and Russian.
- The "Sponsored Discontent" operation, described in OpenAI's February 2025 report, used ChatGPT to generate English social posts and Spanish-language articles, with activity consistent with mainland Chinese business hours.
Context
According to OpenAI, the "Doppelganger" campaign's content primarily focused on the war in Ukraine, portraying Ukraine, the US, NATO, and the EU negatively while presenting Russia positively. The "A2Z" operation leveraged AI to manage fake social media personas, generate bios, analyze posts, draft replies in multiple languages, and proofread them. The "Stop News" operation was notably prolific in its use of DALL·E generated imagery, often in a cartoon style, to attract attention.
Why It Matters
These disruptions highlight the ongoing efforts by AI developers to identify and mitigate malicious uses of their models for influence operations. For builders, these cases underscore the importance of robust safety mechanisms and monitoring for misuse, as well as the potential for AI to both enable and expose such activities.
What To Do
- Note the different tactics, techniques, and procedures (TTPs) observed across these operations.
- Watch for OpenAI's future reports on malicious uses of AI to understand evolving threats.
- Review the OpenAI blog posts to understand the specific indicators of inauthentic activity, such as engagement patterns and account creation dates.