OpenAI revealed that it has banned a group of accounts associated with what it identified as an Iranian covert influence operation that used ChatGPT to create content focusing on topics like the upcoming U.S. presidential election.
In a statement, OpenAI confirmed that it took down a cluster of ChatGPT accounts related to the operation known as Storm-2035.
The operation utilized ChatGPT to produce content covering various subjects, including discussions on U.S. presidential candidates, and distributed this content through social media and websites.
Despite the effort, OpenAI mentioned that the content did not garner significant engagement, with most social media posts receiving minimal interaction. Additionally, there was little evidence of the long-form articles created with ChatGPT being shared on social media platforms.
The articles touched on U.S. politics, global events, and were posted on websites masquerading as both progressive and conservative news outlets, indicating an attempt to target individuals across the political spectrum.
OpenAI stated that their ChatGPT tool was used to generate comments in English and Spanish, which were then posted on multiple accounts across different platforms. Some of these comments were created by rephrasing existing comments from other social media users.
The operation also produced content on various topics such as the Gaza conflict, Israel’s participation in the Olympics, and the U.S. presidential election, interspersed with fashion and beauty-related content to enhance authenticity and appeal.
Last week, Microsoft highlighted Storm-2035 as one of the threat activity clusters mentioned, describing it as an Iranian network engaging U.S. voter groups with polarizing messaging on political issues.
The operation’s fake news and commentary sites included EvenPolitics, Nio Thinker, Savannah Time, Teorator, and Westland Sun, which were observed plagiarizing content from U.S. publications. The group has been active since 2020.
Microsoft also warned about increased foreign influence activities targeting the U.S. election, particularly from Iranian and Russian networks.
As part of their evolving tactics, the propaganda network has started using non-political posts and ads to deceive users and evade detection, by spoofing entertainment and news outlets to redirect users to geopolitical articles.
Based on Meta’s findings, the network has been shifting its strategy to include fewer political posts but with more redirection layers in an attempt to circumvent detection.
