By blocking fake networks from Russia and Iran, OpenAI has unveiled the use of ChatGPT for generating analytical articles and social engineering.
OpenAI, the world's most prominent artificial intelligence company, claims in a new report that a network of users based in Iran and Russia attempted to publish political content related to Iran-US tensions in international media and social networks by using ChatGPT and fake identities. This claim has not yet been verified by any other entity.
The creator of ChatGPT announced that it has identified and blocked the accounts associated with the activities in question; however, it emphasized that it has not been able to determine the exact identity of the operators behind this network.
OpenAI claims that the operators of both operations, by combining artificial intelligence tools with traditional content dissemination methods, attempted to convey their political messages to audiences through seemingly independent identities and entities.
"Operation Scribes"
OpenAI has named the operation attributed to Iranian users "Operation Scribes." According to the company's claims, a set of ChatGPT accounts originating from Iran concealed their geographic location using VPNs. Users of these accounts primarily submitted their prompts in Persian and utilized AI models to generate and edit Persian and English content.
OpenAI states that the operators of this activity used seven journalistic personas who presented themselves as Western, primarily American, writers. Among the names used are Michael Harrison, Sophia Gonzalez, Alice Johnson, and Erwin B. Hoskins.
According to OpenAI, some of these personas had accounts on social media networks including X, Facebook, Instagram, and Threads. For example, one account identified itself as an independent American writer, while social media transparency information indicated the location of the associated accounts was in Iran.
Publication of nearly 100 articles in various media outlets
The most significant part of OpenAI's claim concerns the publication of articles in online media. The company says that by examining public sources, it has identified nearly 100 articles published or republished under the names of the authors attributed to the network across more than 12 media outlets.
The oldest identified article dates back to 2025 and the most recent to October 2026. According to the report, the number of articles increased significantly following the start of the military conflict between Iran and the United States.
OpenAI claims that the users first provided drafts of English articles to ChatGPT for review and editing, and then asked the tool to draft emails tailored to the submission criteria of each media outlet. TahaSite reports.
The topics of the articles mainly concerned Iran-U.S. relations, developments in the Middle East, and the consequences of the war. The company emphasizes that not all published content was necessarily generated by artificial intelligence models, and that part of the use of ChatGPT was dedicated to editing text and preparing publication proposals.
Mass Comment Generation and Attempts to Influence Users
Another part of the report is dedicated to the mass generation of comments on social networks. OpenAI claims that users associated with the network would provide ChatGPT with the text or images of political posts and ask it to generate responses aligned with their desired positions.
OpenAI also states that more than 20 sets of critical Persian responses were generated for posts on social networks. The company acknowledges that the available evidence does not indicate widespread success for this activity on social media. Many of the identified comments received little engagement, and the publication of a large portion of the generated content could not be independently verified.
OpenAI's Actions and the Extent of Alleged Impact
OpenAI has announced that it has blocked the accounts associated with this activity and has shared information regarding the case with relevant authorities and industry partners.
The company assesses the dissemination of articles in the media at level four on a six-degree scale of information influence operations; however, it assesses the network's activity in the social media comments section as level two due to limited engagement.
OpenAI also claims that the network operators used misleading metrics in their internal reports to exaggerate their impact, including attributing the increased views of others' posts to comments they had posted themselves.
Despite these claims, the company clarifies that it has not been able to identify a specific actor behind the operations. According to OpenAI's assessment, the pattern of activity is consistent with a commercial firm providing influence operations services, but the report offers no definitive evidence attributing the management of this network to the Iranian government.
The information released regarding the goals, methods of operation, and scope of these activities is based primarily on OpenAI's own investigations and should be evaluated within the context of the claims and findings announced by the company.
Comments