Imagine scrolling through your news feed and pausing on an article written by an American freelance journalist with a clean, professional headshot, a modest social media footprint, and a byline that looks completely legitimate. The piece is well-written, carefully argued, and published on a real outlet with an established audience. You might share it, cite it, or let it quietly shape your view of a distant geopolitical conflict. But what if that journalist never existed? What if the byline was manufactured in Tehran, the article polished by artificial intelligence, and the entire purpose of the exercise was not to inform you, but to deceive you? That is exactly what OpenAI says it uncovered in a recent investigation into an Iran-based influence operation. Dubbed “Bogus Bylines,” the campaign did something more sophisticated than flooding social media with bot accounts or posting inflammatory memes. It created seven fictitious journalist identities, mostly posing as Americans, and used them to place nearly a hundred articles in online outlets around the world. The operators, working from Iran, used ChatGPT to refine English-language writing, tailor submissions to editors’ guidelines, and compose convincing pitch emails. Some fake personas were reinforced with accounts on X, Facebook, Instagram, and other platforms. One identity claimed to be an American freelance writer, but platform data traced the account back to Iran. The goal was not merely to generate content; it was to launder propaganda through genuine media, inserting fabricated voices into the very institutions that people trust to separate fact from fabrication.
The mechanics of the operation reveal a chilling understanding of how modern journalism works. Editors are overwhelmed with pitches, and a well-written article from a seemingly credible freelancer can easily slip through the cracks, especially if it aligns with a publication’s editorial interests. The Iranian operators were not clumsy amateurs. They studied submission requirements, adapted their writing to fit different outlets, and built social media profiles to create the illusion of a real person with a professional history and personal opinions. Once the articles were accepted and published, the operation’s work was amplified further. Some publishers promoted the planted articles through their own social media accounts, unintentionally lending their credibility to a foreign influence campaign. This gave the operation access to audiences that would never knowingly consume Iranian state propaganda. The timing was telling: after the U.S.-Iran conflict escalated, publication activity surged. The network pushed narratives that portrayed Iran as a resilient victim, the United States as the aggressor, and Israel as an unreliable ally. These were not overt, cartoonish propaganda pieces; they were subtle, persuasive, and designed to fit comfortably within the editorial norms of independent media. OpenAI rated the operation Category 4 on its six-point influence scale, acknowledging the significant threat it posed. Notably, OpenAI did not identify the organization behind Bogus Bylines, but noted that its activity was consistent with a commercial influence-for-hire operation. In other words, the ability to manufacture fake journalists and place propaganda in real publications is not just a tool of the state; it has become a service that can be bought and sold.
But treating this case in isolation would obscure a much larger and far better documented pattern. Only weeks before OpenAI’s report, Anthropic, the company behind the AI model Claude, exposed three Iranian state-aligned operations tied directly to named propaganda institutions of the clerical regime. These included the Islamic Culture and Communications Organization under the Ministry of Culture and Islamic Guidance, the Islamic Propaganda Office of Khorasan Razavi, and the Islamic Propaganda Organization’s Bina Cultural Observatory. According to Anthropic, the operators themselves described their work as “soft war” and “cognitive warfare.” This is not the language of a few rogue hackers; it is the language of a coordinated strategic campaign. The use of AI was systematic and deeply integrated into every stage of the operation. Claude was used to construct influence strategies, create fake personas, build target databases, and draft propaganda manuals. It transformed government intelligence bulletins into multilingual content, disguising official narratives as material from independent or foreign writers. It manufactured apparently grassroots campaigns, making it difficult for even careful observers to distinguish between genuine public sentiment and coordinated manipulation. Some of this material was disseminated through channels aligned with the Revolutionary Guards, Iran’s powerful ideological military force. The sophistication of these operations suggests that Iranian state institutions have fully embraced AI as a force multiplier for their long-standing propaganda efforts. They are not simply experimenting with new tools; they are building professional pipelines that turn raw intelligence and political messaging into polished, plausible, and portable content designed to cross borders and shape opinions.
The same trajectory has been documented by Google, whose Threat Intelligence Group reported in September that Iranian actors used generative AI to create fictitious personas and produce state-aligned counter-influence narratives. Some operators went so far as to instruct AI models to behave like psychological-warfare specialists and to apply persuasive and manipulative techniques in support of regime objectives. This is not a hypothetical concern or an overreaction to isolated incidents. Google’s earlier research found that Iranian government-backed actors were the largest government-linked users of Gemini among the advanced threat groups it tracked. Their targets and interests included foreign governments, defense organizations, and Iranian dissidents. The use of AI allowed these actors to scale their efforts, generating content that was not only more abundant but also more adaptive. Instead of repeating the same talking points, the AI could tailor messages to specific audiences, adjusting tone, language, and argumentation to maximize impact. For Iranian dissidents, this represents a direct threat. The same tools that manufacture fake journalists abroad can be used to identify, target, and harass critics of the regime. For foreign governments and defense organizations, the threat is more diffuse but equally serious: AI-generated propaganda can corrupt open-source intelligence, pollute public discourse, and make it harder for policymakers to distinguish between authentic reporting and manufactured narratives. The consistency of these findings across multiple technology companies, who are competitors and do not ordinarily share intelligence, makes the evidence difficult to dismiss.
Meta, the parent company of Facebook, Instagram, and WhatsApp, has also repeatedly dismantled Iran-based coordinated influence networks posing as Americans. One 2026 network used fake activists, students, and graphic designers supposedly living in U.S. cities, some of whom published AI-generated material. Another Iranian campaign constructed elaborate fake news outlets and personas to push Tehran’s regional interests. These operations are not always sophisticated; sometimes they are crude and easily detected. But the pattern is unmistakable. Iranian actors are constantly probing the defenses of social media platforms, testing new techniques, and adapting to takedowns. The fact that Meta, Google, OpenAI, and Anthropic have all independently identified similar activity suggests that this is not a niche problem but a systemic one. The picture emerging across competing technology companies is therefore difficult to dismiss. It reveals a coordinated, institutionalized effort by the Iranian regime to exploit every available avenue to spread its narratives, undermine its adversaries, and maintain control over its own population. The tools change—from simple Facebook pages to AI-generated personas and fake bylines—but the underlying strategy remains the same: hide the source, invent credible identities, imitate independent journalism, manufacture consensus, and push the regime’s narratives into audiences that would reject them if the real author were visible.
What does this mean for the rest of us? It means that the information ecosystem we rely on is more fragile than we would like to believe. The distinction between authentic journalism and manufactured propaganda is becoming increasingly difficult to draw, and not because the propaganda is crude, but because it is designed to look exactly like the content we have learned to trust. A byline is no longer proof of a person. A publication’s masthead is no longer proof of independence. A polished article with citations and a neutral tone is no longer proof of good faith. The technology is new, but the dictatorship’s method is not. For decades, authoritarian regimes have understood that the most effective propaganda is the kind that does not look like propaganda at all. The best way to convince someone to believe a lie is to make it look like the truth, ideally coming from a source they already trust. Artificial intelligence has not changed that fundamental logic; it has simply made it easier, cheaper, and faster to execute. We are now living in a world where a handful of people in a Tehran office can create an entire newsroom of fictional journalists, write a hundred articles, place them in real publications, and have them promoted by unsuspecting editors. They can target specific audiences, adapt their messages in real time, and do it all with a level of scale and precision that was unimaginable a decade ago. The response cannot be simply to blame social media companies or to demand better AI safeguards, though both are necessary. We need a new kind of vigilance, a willingness to question not just the content but the source, and an understanding that the tools of persuasion have been weaponized. The technology is new. The dictatorship’s method is not. But the stakes have never been higher.