OpenAI said users in Iran used its models to produce about 100 articles under seven fake journalist names, but it could not identify who was behind the campaign.
The articles were published under seven invented bylines and criticised the US war against Iran, according to reports by CNN and the Washington Post. Both outlets cited a report released by OpenAI on Thursday.
OpenAI said the users sent instructions to its chatbot in Persian. CNN reported that the company described the prompts as asking for content that portrayed the United States as "the aggressor" and Iran as "a defiant but resilient victim".
The two outlets gave different figures for how widely the material spread. CNN reported that roughly 100 articles by seven fake journalists appeared in about a dozen outlets worldwide. The Washington Post said OpenAI put the figure at about 100 articles across at least 20 outlets.
The Washington Post reported that one of the bylines, Ervin Hoskins, appeared on an opinion piece in the Port St. Joe Star, a small weekly newspaper in Florida. The piece ran last month and concerned the Iran war. The newspaper quoted in the article said: "The bill will follow Americans beyond the fighting itself." The Port St. Joe Star's response was not seen in the reports reviewed by Tehran Telegraph.
According to the Washington Post, the same byline also appeared in Daily Kos and Middle East Monitor. It said the Daily Kos articles appeared on a community platform that is not subject to review by the site's editors.
The Washington Post said the articles highlighted the negative effects of the war on Americans. CNN reported that they dealt with missile strikes, ceasefires and diplomatic manoeuvring, citing OpenAI.
Middle East Monitor and Daily Kos said they had taken the articles down and were reviewing their processes, the Washington Post reported. Daud Abdullah, the director of Middle East Monitor, told CNN he would look into the flagged articles and "take the appropriate action".
CNN, citing OpenAI, also reported that after the articles were published, the users asked ChatGPT to generate social media comments posing as reactions from readers. NewsBytes, a secondary outlet, reported that OpenAI banned the accounts involved.
OpenAI did not name a state actor. In its report, carried by both outlets, the company said the activity appeared consistent with a commercial actor running a for-hire influence campaign, but that it was unable to identify the particular actor involved.
CNN reported that OpenAI said it was unclear who was behind the campaign and whether they were affiliated with the Iranian government. The Washington Post also noted that it was unclear who was behind some of the apparently fake personas, such as one called Blanchard.
No response from Iranian authorities had been reported in the coverage reviewed by Tehran Telegraph. The origin of the campaign rests on OpenAI's account alone.
Tehran Telegraph has not independently read OpenAI's original report and relies on the accounts of it published by CNN and the Washington Post. The Washington Post has a content partnership with OpenAI.
Daily Kos and Middle East Monitor together have more than 3 million Facebook followers, the Washington Post reported. The articles were therefore placed on platforms with potentially large audiences, although the reports reviewed do not give figures for how many people read the pieces in question.
Photo: Jernej Furman from Slovenia / Wikimedia Commons (CC BY 2.0)
.jpg){kind=link}