OpenAI has announced that it detected and blocked accounts involved in organized operations based in Russia and Iran that sought to influence international public opinion using ChatGPT. The operations included publishing articles in real media outlets under the names of fictional journalists and disseminating political messages under the guise of research institutions, confirming cases in which artificial intelligence (AI) was combined with existing information manipulation tactics.

According to OpenAI's report on "Disrupting deceptive campaigns using AI," accounts operated from Russia and Iran used ChatGPT to write and edit content and to compile internal activity reports. A common feature of the two operations was the attempt to leverage the credibility of media outlets and research institutions while concealing their true sponsors.

Iranian Operation... "Seven Nonexistent Reporters Spread About 100 Articles"

In the Iranian operation, article publication activities were confirmed using the names of seven nonexistent reporters. The operators wrote English-language articles on topics such as the U.S.-Iran conflict, Middle East affairs, and international politics, then polished the text with ChatGPT and revised it to meet each outlet's submission guidelines. They were also found to have written emails proposing the articles to online media outlets.

OpenAI's analysis of publicly available material showed that approximately 100 articles were published or reprinted under these fictional bylines in around a dozen online media outlets worldwide. The articles appeared mainly in small and mid-sized outlets covering international affairs and geopolitical issues. The earliest confirmed article appeared in July 2025, and article publication increased sharply after the start of the U.S.-Iran conflict.

The fake reporters were given careers and areas of expertise to make them appear to be real journalists. Some posed as freelance journalists from the United States and even operated social media accounts to build credibility. However, the accounts' location data revealed connections to Iran.

Attempt to Inflate Opinion Through Comments... "Actual Impact Limited"

The Iranian operation went beyond publishing articles. Operators mass-produced comments in English and Persian for posts about the U.S.-Iran conflict. They also attempted to make it appear that public reaction was substantial by repeatedly posting arguments favorable to certain countries or comments supporting their own published articles.

However, evidence that these comment activities broadly shifted actual online opinion was limited. OpenAI said engagement with the identified comments generally remained in the single or double digits.

Russian Operation 'Dark Clark'... Fronted by Fictional Figure 'Mia Clark'

The Russian operation took a more complex form. Dubbed "Dark Clark" by OpenAI, the operation was assessed as an effort to damage Ukraine's reputation among Latin American countries and to influence domestic politics in some of those nations.

Operators were found to have controlled a research platform called the Social Research Center (SRC), fronted by a fictional figure named "Mia Clark." In this process, there were indications that local residents participated in research and content production without knowing their actual employers or the organization's purpose. OpenAI stated that the organization's internal reports included staff hiring and firing, salary structures, and management of research projects. This amounted to an information operation with the outward appearance of a genuine organization, going beyond simply running fake accounts.

The Russian operation also involved emails impersonating schools and government agencies, manipulated voice recordings, and forged contracts. Some false information spread widely enough to prompt local media coverage and official rebuttals from government agencies.

Rated at Stage 5 of 6 on Spread Scale... "First Since Reports Began"

OpenAI rated the Russian operation at stage 5 on its six-stage spread assessment scale. This is the first case among the operations OpenAI has disrupted since it began issuing such reports to reach stage 5. The Iranian operation's article publication activity was classified at stage 4.

AI's Role Was More 'Distribution and Management' Than 'Production'... Concerns Over Circumventing Verification

A notable point in this report is that AI did not generate all of the false information from scratch. In the Russian operation, ChatGPT was used more for writing internal performance reports than for creating content to spread externally. The Iranian operation also made active use of AI in the content distribution process, such as revising previously written manuscripts to meet media outlets' requirements and drafting submission emails.

This shows that AI can be abused not only to accelerate the production and linguistic polish of information operations but also to circumvent the verification procedures of established media and research institutions. When articles written under fictional bylines are published in real media outlets, readers find it difficult to verify the circumstances behind the content or the identity of the author. These cases also demonstrated that distribution through media outlets can potentially reach a far wider readership than simple fake social media posts.

However, OpenAI noted that there were cases in which the operation operators exaggerated their own results, and the actual spread of some activities could not be confirmed. It is therefore necessary to distinguish between the organizations' claims and verified findings.

"Natural-Sounding Prose Alone Is Not Enough... Verify Authors' Existence and Sources"

These cases raise the problem that, in the era of generative AI, judging the credibility of news content solely by the naturalness of the writing or the completeness of the article format is insufficient. The need is growing to verify whether the person who wrote an article actually exists, the origin of the manuscript, and the interests of the organization behind it.

OpenAI stated that the purpose of disclosing the details of these operations is to aid follow-up research and disruption efforts and to make it harder for similar information operations to continue.

[This article is based on OpenAI's report on "Disrupting deceptive campaigns using AI."]