In a move that underscores the growing intersection of artificial intelligence and geopolitical influence, OpenAI has lifted the lid on two sophisticated influence operations that ran on its platform. The company’s threat intelligence teams revealed they had been tracking a sprawling Russian operation, dubbed “Dark Clark,” and an Iranian campaign, known as “Bogus Bylines,” which had been using ChatGPT to generate content, craft fake personas, and plant stories across the internet. What made these campaigns noteworthy wasn’t just their brazenness, but the fact that they were able to break through into real-world news cycles, prompting responses from politicians and drawing the attention of fact-checkers. According to OpenAI’s analysis, the Russian operation was assessed as a Category 5 threat—the highest level it uses to describe attempts to run front organizations—while the Iranian campaign was broken down into two parts, with its article-planting workstream rated at Category 4. Both efforts leaned on the AI assistant not just to write, but to translate, refine against specific media outlets’ submission guidelines, and even generate pitch emails, marking a new, more accessible phase in the business of influence.
The Russian operation, which OpenAI describes as the most complex attempt to run a front identity it has disrupted in the past two and a half years, was centered on a fake think tank called the Social Research Center. The group’s operatives used ChatGPT to manage their activities, producing internal reports, drafting contracts, and even asking the model to write operational content such as fake letters implicating Ukraine’s honorary consul in Panama in corruption. The SRC’s website claimed to focus on the Indian diaspora in Latin America, and the people it hired on the ground were apparently unaware they were working for a Russian group. The operation’s targets were largely in Latin America, and the impact was tangible: fabricated audio of a worker at Bolivia’s state water company warning of a shutdown, false claims about Ecuador’s president insulting poor people, and a fake contract for infrastructure services that was circulated on social media to back up a story about Ukrainian recruitment. OpenAI noted that the operators often exaggerated their success, even asking ChatGPT to identify incidents they could take credit for despite having no involvement, which suggested the entire effort was designed to impress their own employers back home rather than to change any real-world audience.
The Social Research Center itself was a curious creature. Its website published over 60 original-looking articles, its X account was registered with a German location but connected from the Russian Federation App Store, and one of its Instagram accounts was administered from Venezuela. A LinkedIn page for the organization claimed 200 to 500 staff and had nearly a thousand followers. OpenAI’s investigation showed that the operators were making decisions about hiring and firing, pay scales, and staffing plans at the SRC, which indicated they controlled the entity completely. The on-the-ground employees, many of whom were likely hired through legitimate-looking job posts, were unknowingly part of a Russian influence operation. The operational content the operators asked ChatGPT to produce was a mix of the mundane and the outlandish: translation of scripts, proofreading of fake contracts, and even a letter using the logo of a Panamanian news channel. While the operation did manage to get some of its content picked up by real media, the engagement was low, with many of the social media likes and comments coming from accounts linked to the operation itself.
On the other side of the spectrum, the Iranian campaign, “Bogus Bylines,” was a more conventional but still successful attempt at planting articles. The operators created seven fake journalist personas, including Ervin B. Hoskins, Noah Lamington, and a certain Michael Harrison, whose biography was written by ChatGPT itself. Interestingly, Meta had previously disrupted an Iranian operation featuring a persona with the same name, and Instagram had banned the Hoskins persona’s account in August 2026. Using these personas, the operators managed to get nearly 100 articles published or syndicated across roughly a dozen small to medium online outlets focused on international affairs, with the earliest appearing in July 2025 and the frequency increasing after the outbreak of the US-Iran conflict. One outlet that published their work had nearly two million followers on Facebook alone. The articles were carefully pitched to editors using emails that ChatGPT helped generate, and the operation also ran a parallel commenting campaign, producing batches of hostile Persian-language replies to UK-based satellite network Iran International and praising their own articles to make them look more popular.
OpenAI’s assessment of the Iranian operation was more nuanced. The commenting workstream was rated at Category 2, reflecting its low engagement and limited impact, while the article-planting workstream was rated at Category 4. The company noted that the operators’ internal reports measured their impact by the number of views on the posts they replied to, rather than views of their own replies, a calculation that greatly exaggerated their apparent effectiveness. Overall, the activity appeared to be the work of a commercial actor running a for-hire influence campaign, though OpenAI said it was unable to identify the specific actor involved. The company shared its findings with industry partners and relevant authorities, who are in a better position to assess the full scope of the engagement. For its part, OpenAI has said it will continue to monitor these operations and use the information to strengthen its defenses. The takeaway is clear: as AI tools become more powerful and accessible, they are being weaponized in increasingly subtle and persistent ways to try to shape public opinion and sow confusion.