OpenAI Bans Russian and Iranian Influence Networks: 7 Fake Journalists and 100 Articles

Two large-scale influence operations exposed

OpenAI has uncovered and blocked two coordinated influence operations: the Russian operation "Dark Clark" and the Iranian operation "Bogus Bylines." Unlike classic social media campaigns, both relied on a strategy of embedding fake content directly into legitimate media sources, maximizing credibility and reach. Artificial intelligence was used mainly for internal reporting and for adapting propaganda content to different languages and regions, allowing for rapid localization of materials for target audiences.

Russian campaign "Dark

Clark": disinformation against Ukraine

The Russian operation "Dark Clark" launched a targeted influence campaign in Latin America aimed at discrediting Ukraine and destabilizing the political situation in the region. The operation was built around control of a think tank run through a fictitious persona. According to OpenAI's assessment, the team involved local staff in the organization's work without their awareness of its true affiliation and real goals.

The operation synthesized and distributed fake audio files, fabricated documents, and disinformation materials. These fakes triggered official denials from governments and drew active attention from fact-checkers in Ecuador and Peru, creating the appearance of genuine public resonance and provoking a political response. The scale of the campaign is significant enough that OpenAI assigned it a rating of 5 out of 6 on its own Breakout Scale for evaluating the effectiveness of influence operations. This is the first recorded case of reaching the fifth category in the past two and a half years. Such a rating is only given when policymakers in the targeted countries respond to the distributed content at the state level.

Iranian campaign "Bogus

Bylines": a network of seven fake journalists

The Iranian operation "Bogus Bylines" used a network of seven fake journalist identities to place nearly 100 articles on the topic of the US-Iran conflict in online media outlets around the world. The strategy was designed to penetrate legitimate news publications, which gave the disinformation the appearance of official information and boosted its credibility.

In addition to publishing articles, the operation generated additional comments and discussions on social media to create the appearance of organic public interest. However, the social media component of the campaign showed minimal effectiveness and gained virtually no traction, unlike the direct media publications.

The role of AI and measures taken

Both campaigns used large language models and AI tools, but primarily for internal operational purposes: compiling and analyzing reports on campaign results, and adapting propaganda content to the linguistic and cultural characteristics of target regions. The extent of direct AI use for automated content generation was limited.

OpenAI blocked all identified ChatGPT accounts linked to both operations, cutting off the perpetrators' access to the tools. The company also notes that it had previously uncovered another Russian campaign built on the use of social media and a fictitious think tank as infrastructure for spreading disinformation.