The operation began with Persian-language prompts to ChatGPT, according to OpenAI, which said users in Iran generated English-language articles, then asked the system to reshape them for particular publications and draft pitch emails to editors. The company described the effort as a commercial, “for-hire influence campaign,” rather than identifying a specific operator.¹

Its reported target was the political fallout from U.S. military action in Iran. OpenAI said roughly 100 articles, often stressing the war’s costs to Americans, were placed at at least 20 outlets under seven fabricated bylines. One was “Ervin Hoskins,” whose article in the Port St. Joe Star warned that “The bill will follow Americans beyond the fighting itself.”¹

For David Adlerstein, publisher of the small Florida weekly, the column did not arrive as an obvious foreign-information operation. It was forwarded by PeaceVoice, an Oregon liberal think tank that had supplied material to the paper for years. Adlerstein said he found the piece “thought-provoking” and less overtly anti-Trump than other PeaceVoice submissions, so he did not verify either the writer or the article’s origin.¹

That trust chain is central to the case. PeaceVoice founder Tom Hastings said the article also looked authentic to him and that he saw no unusually pro-Iran message—only “a legitimate critique of Trump’s war of choice.” After OpenAI contacted the group, PeaceVoice strengthened its no-AI policy; Hastings said future contributors may be required to meet by Zoom.¹

Other outlets have moved more decisively. Middle East Monitor said it removed articles attributed to the fake authors, while OpenAI said it banned the accounts involved, despite already blocking ChatGPT access from Iran. The company says it shared details with authorities and published its findings to expose vulnerabilities that influence operators seek to exploit.¹