OpenAI Disrupts AI-Enabled False Front Influence Operations
OpenAI has disrupted two sophisticated influence operations (IO) originating from Russia and Iran that utilized AI to establish "false front" entities. These operations moved beyond simple social media botting to create fake journalist personas and a fraudulent think tank, successfully planting geopolitical messaging into mainstream media outlets.
The Role of AI in False Front Operations
AI is being used by threat actors to increase the scale, efficiency, linguistic fluency, and editorial quality of deceptive campaigns. Rather than replacing traditional influence tactics, AI is used to streamline workflows, such as refining long-form articles to meet specific publication criteria or drafting internal performance reports.
OpenAI notes that these operations closely resemble pre-AI era influence campaigns but leverage AI to exploit unsuspecting editors and employees, making it easier to plant content in front of audiences unaware of the source's motivations.
Operation "Dark Clark": Russian Influence in Latin America
OpenAI banned a cluster of Russian-origin accounts that targeted Latin American countries, specifically aiming to undermine Ukraine's reputation and influence political outcomes in Argentina, Bolivia, and Ecuador.
The Social Research Center (SRC) Front
The most complex element of this operation was the creation of the Social Research Center (SRC), a fake "research platform" focusing on the Indian diaspora in Latin America. The Russian operators controlled the SRC via a fake persona named "Mia Clark."
Crucially, the operation co-opted unwitting local employees in Latin America who believed they were working for a legitimate entity. These employees produced original research, which the Russian operators then used to build a durable front identity. This differs from previous operations that relied primarily on plagiarized content.
Tactical Execution and Impact
The operators used ChatGPT primarily for internal reporting to superiors and occasionally for creating fakes. Key activities included:
- Fabricated Events: Tricking schools in Peru into holding events featuring controversial figures to trigger diplomatic tensions between Ukraine and Poland.
- Fake Documents and Audio: Creating fabricated audio of Ukraine's consul in Ecuador and a fake contract to suggest the Ecuadorian government was recruiting citizens to fight in Ukraine.
- Internal Deception: Attempting to take credit for existing political arguments or news events (such as the Falkland/Malvinas islands dispute) to inflate their effectiveness to their own employers.
OpenAI assesses this operation as Category 5 on the Breakout Scale, the first Category 5 operation it has disrupted, due to its ability to provoke official government denials and public comments from politicians.
Operation "Bogus Bylines": Iranian Influence Campaign
OpenAI disrupted an Iranian operation that used AI to seed long-form articles in online news outlets under deceptive bylines.
The Journalist Persona Strategy
The operation utilized seven fake Western journalist personas (e.g., "Ervin B. Hoskins" and "Sophia Gonzalez") to pitch articles to small and medium online outlets. The actors used ChatGPT to:
- Review and refine English-language drafts to match the submission criteria of specific outlets.
- Generate professional pitching emails to editors.
- Write detailed biographies for the fake personas.
OpenAI identified nearly 100 articles published under these bylines across a dozen publications, primarily focusing on the US-Iran conflict and portraying the US as an aggressor.
Social Media Amplification
Alongside the articles, the actors generated batches of social media comments in English and Persian to praise their own articles or attack critics of the Iranian regime. However, these comments generally saw low engagement.
Impact Assessment
OpenAI applied a split assessment using the Breakout Scale:
- Journalist Personas: Rated Category 4 due to consistent breakout into multiple media outlets, some with millions of followers.
- Social Media Replies: Rated Category 2 due to low engagement and a lack of ability to dominate conversations.
Summary of Influence Operation Reach
Analysis of 30 covert influence operations exposed by OpenAI over the last two and a half years reveals a clear pattern: operations that successfully land content in real media outlets have a significantly higher potential reach and impact than those relying solely on fake social media distribution.