OpenAI said Thursday that a series of opinion pieces critical of the Iran war, published under fake bylines in U.S. outlets, was generated with its ChatGPT chatbot by users based in Iran as part of an "influence campaign." The campaign placed about 100 fake articles across at least 20 news outlets, the company said. One byline, Ervin Hoskins, appeared in the Port St. Joe Star, a weekly newspaper in Florida, as well as on Daily Kos and Middle East Monitor, which together have more than 3 million Facebook followers. The author photo used by the Star was also generated with OpenAI's tools, according to the company's verification tool.
The Star's publisher, David Adlerstein, said the column was pitched by Peace Voice, a small liberal think tank in Oregon that had sent the paper content for years. Because of that relationship, he did not verify the author's identity. Tom Hastings, founding director of Peace Voice, said the group strengthened its no-AI policy after OpenAI contacted him.
OpenAI said it banned the accounts behind the campaign and blocks ChatGPT access from Iran. The company said the activity was consistent with a commercial actor running a for-hire operation, but it could not identify the actor and had shared information with the relevant authorities. Middle East Monitor and Daily Kos removed the articles. Daily Kos founder Markos Moulitsas called the campaign "incredibly corrosive" to his platform, where community posts are not reviewed by staff before publication.
The episode follows earlier cases in which Business Insider and Wired removed AI-generated articles under the fake byline Margaux Blanchard. Such incidents show how AI tools can help fabricated personas pass editorial checks, particularly at smaller outlets that rely on outside submissions.
Source: Hacker News · Summarized by HeadlinesBriefing