Blocking Iran's Covert Influence Operation

OpenAI detected and blocked accounts linked to Storm-2035, a covert influence operation by Iran that used ChatGPT to generate content on various topics, including the US presidential campaign, in an attempt to manipulate public opinion.

The operation was run in two main ways:

• Long-form article generation: Articles on US politics and global issues were posted through 5 websites disguised as progressive and conservative-leaning news outlets.
• Social media comments: Short comments in English and Spanish were generated and posted on X (formerly Twitter) and Instagram. Some accounts also asked the model to rewrite other users' comments.

The main topics of the generated content included the Gaza conflict, Israel's participation in the Olympics, and the US presidential election, with fashion and beauty-related comments mixed in to boost account credibility or gain followers.

The actual impact of this operation is assessed to be minimal. According to Brookings's Breakout Scale, the operation remained at the low Category 2 level, with little to no spread among real users. OpenAI is sharing the threat intelligence gathered from this case with government and industry stakeholders to strengthen its response.