OpenAI just shared news about taking down an Iran-connected influence campaign called Storm-2035, which was making content about the U.S. presidential election. This group used a bunch of ChatGPT accounts to create AI-written articles and social media posts to try to sway public opinion.
The influence campaign focused on making content on different subjects, including thoughts on U.S. presidential candidates. To seem more authentic, the bad actors also made content about things like the Gaza conflict, Venezuelan politics, and even fashion and beauty topics.
The operation spread its content on five websites pretending to be progressive and conservative news sources. These sites were niothinker.com, savannahtime.com, evenpolitics.com, teorator.com, and westlandsun.com. The content created then made its way to social media platforms like X (Twitter) and Instagram, aiming at both English- and Spanish-speaking readers.
OpenAI stated that the campaign’s poor engagement didn’t connect with actual users. Brookings’ Breakout Scale, which measures how much covert influence operations affect people, put this campaign near the bottom of Category 2. This rating shows that even though the operation used several platforms, real users didn’t pay much attention to it or share it a lot.
Although the operation didn’t achieve much, OpenAI took this seriously. After shutting down the accounts, the company shared what it had found with the government, campaign, and industry partners to strengthen joint efforts against such operations. OpenAI stays committed to fighting abuse on a large scale, using the power of generative AI to disrupt foreign influence campaigns and safeguard democratic processes.
