OpenAI says it has banned accounts linked to covert influence operations originating in Russia and Iran, providing another example of how generative artificial intelligence is being incorporated into political propaganda networks. According to the company, the campaigns used AI tools to produce or refine content, but combined those capabilities with older tactics including fake journalist identities, fabricated research platforms and coordinated publication efforts. The disclosure comes as governments and technology companies prepare for election-related information threats and broader geopolitical competition.
The Russian-linked campaign focused on Latin America
OpenAI said one operation connected to Russia used a purported research platform that published material aimed at audiences in Latin America. The network presented content in formats designed to look like independent analysis or journalism rather than state-aligned messaging.
AI tools were used to help generate text and support content production. OpenAI said the campaign still depended heavily on traditional influence techniques, suggesting generative AI acted as an efficiency tool rather than an entirely new propaganda method.
The Iranian operation used fake journalists
The Iranian-linked campaign created seven fabricated journalist identities and used them to pitch articles to online publications. Nearly 100 pieces were reportedly published under those names at more than a dozen outlets.
OpenAI said publication activity increased significantly after the outbreak of the U.S.-Iran conflict. The use of fake bylines is important because it allows influence operators to place material inside legitimate or semi-legitimate media ecosystems rather than relying only on anonymous social media posts.
AI lowers the cost of producing believable content
Influence campaigns have existed long before generative AI. What changes with modern language models is the speed and cost of producing large amounts of readable material in multiple languages. Operators can draft articles, translate messages, adjust tone and generate variations for different audiences much faster than before.
That capability can make relatively small teams look larger and more professional. It can also help campaigns adapt messages to local political language without requiring the same number of native speakers or editors.
But distribution remains harder than generation
OpenAI and other technology companies have repeatedly noted that AI-generated propaganda does not automatically become influential. Creating content is easy; getting real people to read, trust and share it remains difficult.
Successful influence operations often depend on existing communities, trusted accounts, media relationships or paid distribution. The latest cases show operators trying to overcome that problem by building fake journalistic identities and placing material on established websites.
Detection depends on platform cooperation
Removing accounts from one AI service does not necessarily end an operation. The same group can use other models, local software or human writers. That means companies need to share indicators with social networks, email providers, cybersecurity firms and governments.
OpenAI publishes periodic threat reports describing how malicious actors use its systems. Transparency can help researchers understand patterns, but disclosures also need enough technical detail to allow independent verification without revealing detection methods that attackers could immediately evade.
Russia and Iran have long histories of information operations
Western intelligence agencies have accused Russia of running influence campaigns around elections, military conflicts and public trust in institutions. Iran has also been linked to operations targeting audiences in the United States, Middle East and other regions.
Moscow and Tehran have frequently rejected or challenged Western allegations, arguing that accusations are politically motivated. Attribution in cyber and influence operations can be difficult because actors use intermediaries, fake infrastructure and compromised accounts.
Election security is an obvious concern
The United States is approaching midterm elections, and many countries have major votes scheduled over the next year. AI-generated political content could be used to impersonate candidates, fabricate news or flood social platforms with persuasive messages.
Deepfake video and audio receive much of the public attention, but text-based influence can be less visible and easier to scale. A fake reporter or research group may appear credible long enough to insert narratives into real political debate.
AI companies face pressure from both sides
Governments want technology companies to prevent misuse, while free-speech advocates warn against opaque moderation systems that could suppress legitimate political expression. Companies therefore need standards that distinguish coordinated deception from ordinary advocacy, satire or controversial opinion.
OpenAI’s actions focused on accounts tied to covert networks rather than users expressing pro-Russian or pro-Iranian views. That distinction is central to maintaining political neutrality while enforcing platform rules against deceptive operations.
The larger issue is authentication
As generative tools become widely available, the internet will contain more synthetic text, images and video. Detecting whether something was produced by AI may become less useful than verifying who published it and whether that identity is authentic.
Media outlets may need stronger contributor verification, while social platforms could expand provenance tools and account-history signals. The fake-journalist tactic described by OpenAI shows why identity verification matters.
AI has not replaced old propaganda—it has upgraded it
The latest disclosure suggests a pattern that security researchers increasingly emphasize: artificial intelligence is being layered onto familiar influence methods rather than replacing them. Fake organizations, misleading identities and political narratives remain the core strategy.
The technology makes those tactics faster and cheaper. OpenAI’s account bans may disrupt the specific networks it identified, but the broader challenge will continue across the AI industry. The key question is whether platforms can make coordinated deception expensive enough that scale no longer provides attackers an easy advantage.
Sources
- OpenAI threat-intelligence disclosure
- Anadolu Agency reporting, Oct. 10, 2026
- Public cybersecurity assessments on foreign influence operations
