August 19, 2024 at 04:20PM
OpenAI banned Iranian-linked ChatGPT accounts for spreading fake news on US presidential campaign, using long-form articles and social comments on Democrat Kamala Harris and Republican Donald Trump. OpenAI identified 12 accounts on X and one on Instagram involved in this covert influence operation. Storm-2035 was attributed to the phony posts, with minimal audience engagement.
From the meeting notes, it is clear that OpenAI has taken action against ChatGPT accounts linked to an Iranian crew suspected of spreading fake news about the upcoming US presidential campaign on social media platforms. The accounts utilized ChatGPT to produce articles and social media comments about both Democratic candidate Kamala Harris and Republican candidate Donald Trump.
Despite being shared on various websites, OpenAI’s alert indicates that this influence operation did not effectively engage with the audience. This covert operation is attributed to the Tehran-backed group Storm-2035, also highlighted by Microsoft for meddling in elections and potentially inciting violence.
Furthermore, Google’s threat hunters recently disclosed intelligence on Iranian cyber influence activity following an increase in attacks resulting in data leaks from the Trump re-election campaign.
OpenAI identified 12 accounts on X and one on Instagram involved in this covert influence operation. The fabricated news and comments covered diverse topics, including the US presidential election, the conflict in Gaza, Israel’s participation in the Olympics, and other politically charged subjects.
The organization stated that the covert operation involved two workstreams, with the first producing articles on US politics and global events published on websites masquerading as progressive and conservative news outlets. The second workstream generated short comments in English and Spanish posted on social media.
Five domains were identified as part of this operation: niothinker[.]com, savannahtime[.]com, evenpolitics[.]com, teorator[.]com, and westlandsun[.]com. Storm-2035 has been active since the US midterm elections in 2022, publishing approximately 10 “articles” per week across multiple languages.
Despite being written in English, Storm-2035 also operates influence-operation websites in Arabic, French, and Spanish. However, based on the Brookings’ Breakout Scale, Storm-2035 only garnered a scale rating of 2, indicating its limited effectiveness in engaging with real individuals on multiple platforms.
OpenAI reported that the majority of identified social media posts received minimal engagement, with few or no likes, shares, or comments, and observed a lack of evidence indicating widespread sharing of the web articles across social media platforms.
Earlier this year, OpenAI closed five accounts utilized by government-backed groups in China, Iran, Russia, and North Korea to generate phishing emails and malicious software scripts.