The Iranian operation, nicknamed “Bogus Bylines” by investigators, operated by using prompts written in Persian to instruct ChatGPT to generate political commentary under seven fabricated author identities. These fake personas—posing as freelance journalists, policy analysts, and peace advocates—pitched long-form articles focusing on the consequences of U.S. military action against Iran to editors at small and medium-sized publications worldwide.
According to OpenAI, the campaign successfully placed approximately 100 articles across at least 20 real news outlets, with the earliest submissions dating back to July 2025. The operation bypassed editorial safeguards by infiltrating syndication networks like PeaceVoice, which supplied commentary to various local publications over the years. One prominent instance involved an opinion piece attributed to a fabricated author named Ervin Hoskins, which was published by the Port St. Joe Star, a small Florida newspaper.
Editorial Vulnerabilities and Syndication Reliance
David Adlerstein, publisher of the Port St. Joe Star, explained that because the submission arrived through a trusted syndication relationship with PeaceVoice, the publication did not independently verify the author’s identity. Following disclosures by OpenAI, Adlerstein reported considering explicit internal policies governing AI-generated submissions. Similarly, the U.K.-based Middle East Monitor confirmed it removed articles published under the fictitious identities flagged by the AI firm, while PeaceVoice strengthened its existing rules against artificial intelligence contributions.
Darren Linvill, co-director of Clemson University’s Watt Family Innovation Center Media Forensics Hub, noted that generative technology provided malicious actors with an unprecedented capability to scale disinformation. “AI was valuable to them above and beyond just being a shortcut,” Linvill said, pointing out that it helped produce language that looked, felt, and read authentically enough to pass editorial scrutiny at legitimate publications.
Rating Scales and Broader Implications
OpenAI rated the Iranian article-planting effort a Category 4 on its six-point Breakout Scale, which measures the spread and impact of disinformation campaigns. In contrast, the network’s social media amplification efforts received only a Category 2 rating, generating single- or double-digit engagement. This disparity highlighted a core trend observed across covert operations tracked by OpenAI: campaigns that successfully breach established media outlets achieve significantly wider distribution than those relying solely on social media channels.
Alongside the Iranian network, OpenAI also disrupted a separate Russian operation dubbed “Dark Clark,” which targeted Latin America to undermine support for Ukraine. That campaign achieved a Category 5 rating after successfully provoking public statements and official responses from politicians and government figures in countries such as Ecuador and Bolivia. OpenAI confirmed it has banned all user accounts associated with both networks and shared its investigative findings with relevant authorities.
Reporting source: sundayguardianlive.com.