OpenAI has taken action against five ChatGPT accounts that were generating content for Iranian influence operations. This activity was detailed in OpenAI’s February 2025 report. The accounts produced articles and social media posts for entities associated with the International Union of Virtual Media (IUVM) and STORM-2035, two operations previously reported by other organizations.
Key Points
- OpenAI banned five ChatGPT accounts for generating content for Iranian influence operations.
- The content was linked to IUVM and STORM-2035 operations.
- One banned account generated content for both IUVM and STORM-2035, suggesting a potential previously unreported connection.
- Content included long-form articles for websites like al-sarira[.]com and lalinearoja[.]net, and tweets for associated X accounts.
- One operator used OpenAI models to generate French-language texts for critiquepolitique[.]com and English-language articles for iuvmpress[.]co.
- An operator rephrased AI-generated content before publication, which OpenAI's models identified as highly likely rewrites.
- The influence operations' content was typically pro-Palestinian, pro-Hamas, pro-Iran, and opposed to Israel and the United States.
Context
According to OpenAI, the banned accounts generated a range of content for various online entities connected to previously identified Iranian influence operations. Four of the five accounts produced content published by entities linked to STORM-2035, while the fifth generated content for both STORM-2035 and IUVM. This overlap suggests a potential, previously unrecorded relationship between the two operations at the operator level.
Why It Matters
This action demonstrates OpenAI's ongoing efforts to detect and disrupt malicious uses of its AI models for influence operations. The findings highlight the evolving tactics of state-linked actors, including the use of AI for content generation and attempts to evade detection through rephrasing.
What To Do
- Note that OpenAI's February 2025 report provides details on specific influence operations.
- Watch for updates from OpenAI regarding their efforts to counter malicious AI use.
- Review the Breakout Scale for assessing the impact of influence operations.