In this photo illustration, the homepage of OpenAI’s ChatGPT app is displayed on a laptop screen on February 3, 2023. This week, OpenAI announced that it had used its own AI models to detect covert influence operations.
Leon Neal/Getty Images Europe
Hide caption
Toggle caption
Leon Neal/Getty Images Europe
OpenAI announced that it had banned a group of ChatGPT accounts after finding evidence that the users had ties to Iranian groups seeking to sow division among US voters.
OpenAI said in a statement on Friday that it found online articles and social media comments created with the help of ChatGPT. The content was not widely disseminated, but it focused on a number of divisive issues, including the U.S. presidential election, Israel’s war in Gaza and Israel’s participation in the Olympics. The AI-generated content catered to both progressive and conservative audiences.
“This activity does not appear to have garnered meaningful audience engagement: the majority of the social media posts we identified received few to no likes, shares or comments,” OpenAI said.
An investigation found that the AI-generated long-form articles were published on websites posing as news organizations, while OpenAI said the now-banned accounts had asked ChatGPT to rewrite comments already posted by other social media users, creating social media comments in both English and Spanish.
“They mixed political content with commentary on fashion and beauty, possibly to appear more authentic or to gain more followers,” the company said.
OpenAI’s internal investigation comes a week after a report from Microsoft’s Threat Analysis Center showed that groups linked to the Iranian government used a range of online tactics in an attempt to interfere in the US presidential election.
The day after Microsoft released its report, President Donald Trump’s campaign said it had been hacked and accused Iranian hackers of stealing classified internal documents. The campaign did not provide specific evidence. Google’s threat analysis group said Wednesday it had detected and blocked phishing attacks by Iranian hackers targeting both the Trump and Biden-Harris campaigns.
What is the motivation behind the Iran-linked operation?
OpenAI said the accounts were linked to a covert operation known as Storm-2035. Last week, Microsoft researchers identified the same group as behind four websites posing as US news organizations.
The fake sites spread polarizing messages about hot-button issues like LGBTQ rights and the conflict between Israel and Hamas in an effort to sow confusion and polarize American voters ahead of Election Day, according to Clint Watts, general manager of Microsoft’s Threat Analysis Center.
“Iran is also focused on preventing the elections from taking place,” Watts told NPR last week.
Both Microsoft and OpenAI reported that their websites and AI-generated content did not receive many page views or online engagement.
