OpenAI Exposes Sophisticated AI-Powered Disinformation Networks
OpenAI disclosed on Thursday that it had successfully disrupted two separate influence operations originating from Russia and Iran, both of which leveraged ChatGPT and other generative AI tools to fabricate journalist personas and establish covert think tanks aimed at manipulating Western media narratives. The company detailed how these campaigns used AI-generated content to seed false stories across legitimate news outlets, prompting fact-checks and official denials from governments and media organizations in multiple countries.
Russian Campaign: Dark Clark and the Social Research Center
One of the operations, designated "Dark Clark" by OpenAI researchers, was attributed to Russian actors and centered on influencing political discourse in Latin America. The campaign revolved around a fabricated entity called the Social Research Center (SRC), presented as a legitimate think tank based in the region. At the heart of this deception was an AI-generated persona named Mia Clark, whom the group portrayed as the SRC’s leader.
According to OpenAI, the SRC was not merely a passive front but an active instrument of Russian influence, with evidence suggesting that the group was in direct control of the organization. This conclusion was drawn from internal ChatGPT-generated reports in which operatives discussed operational details such as employee wage scales, hiring, and firing decisions — indicators that pointed to operational authority rather than passive observation.
OpenAI described the Dark Clark operation as "the most complex attempt to run a front identity that we’ve disrupted over the past two and a half years." The company noted that individuals working for the SRC in Latin America appeared unaware of their connection to Russian actors, distinguishing this campaign from other Russia-linked think tanks previously exposed, where co-opted staff were more likely to be knowingly complicit.
The SRC’s content primarily sought to damage Ukraine’s reputation across Latin America, but also extended into localized political issues in Argentina and Bolivia. OpenAI cited multiple instances where the group’s disinformation triggered official responses: in March, Lupa Media debunked a TikTok video falsely claiming the Ecuadorian government was using shell companies to recruit citizens to fight for Ukraine against Russia; the same network produced fake audio clips of the Ukrainian consul in Ecuador allegedly insulting Ecuadorian citizens. In May, during anti-government protests in Bolivia, the group circulated a fabricated audio clip of a state water company employee claiming the government was about to declare a state of emergency and cut off water to La Paz — a claim promptly denied by Bolivian authorities.
While OpenAI did not attribute the Dark Clark campaign to any specific Russian government or intelligence agency, it highlighted the operators’ pronounced interest in Politology, a successor organization to the Wagner Group. The firm reported that prompts referencing Politology appeared "far more" frequently in the group’s interactions with AI models than references to any other Russian entity.
Iranian Campaign: Fake Journalists and Limited Engagement
The second operation, linked to Iranian actors, employed similar AI tools to generate written articles focused on the U.S.-Iran conflict. Operators created at least seven distinct fake journalist personas who managed to place dozens of stories through legitimate online media outlets.
OpenAI reported identifying nearly 100 articles published or syndicated under these fabricated bylines across approximately a dozen online outlets worldwide. These outlets were generally small to medium-sized and specialized in international affairs, geopolitics, and Middle Eastern events. Despite the volume of placement, the stories often failed to generate significant audience engagement, receiving few comments or social media reactions.
The company characterized the Iranian effort as resembling "a commercial actor running a for-hire influence campaign," though it declined to attribute the operation to any specific Iranian state or non-state actor.
Impact Assessment and Broader Context
OpenAI evaluates influence campaigns using an impact scale ranging from 1 to 6, with most tracked operations scoring at the lower end due to minimal organic reach and negligible offline effects. In this case, the Russian campaign was rated a 5, while the Iranian campaign received a 4 — scores indicating a notable ability to break through to mainstream audiences and provoke institutional responses such as fact-checks and governmental denials.
These ratings suggest that both campaigns succeeded in generating narratives substantial enough to compel official rebuttals, a threshold typically reached only when disinformation gains widespread traction. OpenAI emphasized that such outcomes remain rare among the influence operations it monitors, underscoring the heightened sophistication and effectiveness of these particular AI-enabled efforts.
Source: CyberScoop