On Thursday, OpenAI confirmed what sounds like a plot pulled from a spy novel, but is now very much part of our real digital world. The company said it had disrupted two covert influence operations, one rooted in Russia and the other in Iran, both of which abused AI tools to create what OpenAI called “false front entities” meant to promote political narratives. In Iran’s case, a network of operators used artificial intelligence to generate social media comments, invent at least seven fake journalist personas, and pitch polished long-form articles to global media outlets. These fabricated writers were given believable Western names, made to look like seasoned reporters, and then deployed to shape conversations around the Russia-Ukraine war and the U.S./Israel-Iran conflict. OpenAI acknowledged that influence operations are nothing new, but artificial intelligence has made them drastically easier to carry out with more speed, more linguistic polish, and more convincing editorial output. There is something deeply unsettling about this. The same everyday technology people use to write emails, brainstorm ideas, or translate languages was quietly turned into a machine for manufacturing false credibility. OpenAI also noted that both campaigns relied on “questionable or outright deceitful methodologies to exaggerate the operators’ effectiveness,” meaning the people running these operations were not only trying to manipulate audiences, but also likely trying to impress whoever was paying them.
The mechanics of the Iranian operation are a stark reminder of how sophisticated these schemes have become. According to OpenAI, a cluster of ChatGPT accounts traced to Iran was used to generate content in both Persian and English. The actors reportedly accessed the tool through a VPN and used it to strengthen long-form articles and editorial pitches that were then published under entirely fabricated bylines. Those bylines belonged to Ervin B. Hoskins, Noah Lamington, Sophia Gonzalez, Michael Harrison, Ericka Feusier, Jenny Williams, and Alice Johnson. On paper, they appeared to be Western journalists covering international politics, the Middle East, human rights, or conflict. The deception was not clumsy. In one example, an article by “Noah Lamington” appeared in International Policy Digest under the headline “The Knesset Revolt Netanyahu Couldn’t Prevent,” a piece designed to portray political instability inside Israel. Another article, this one by “Sophia Gonzalez,” was published in the News Register under the headline “Sophia Gonzalez: When diplomacy becomes another weapon of war.” A third persona, “Ericka Feusier,” appeared to write extensively about insecurity in Iraq and American unemployment, with work popping up on websites such as Eurasia Review, Common Dreams, and the IA Forum. OpenAI identified nearly 100 articles produced by this operation, spread across syndicated publications and small- to medium-sized media outlets around the world. The timeline matters too: the earliest article appeared in July 2025, and the latest surfaced in October 2026, with output increasing sharply after the outbreak of the U.S.-Iran conflict. This was not a one-off experiment. It was a sustained, evolving propaganda project that adapted to real-world events.
What makes this strategy so dangerous is the way it borrowed trust from real media outlets. One online publication that ran articles from this influence campaign had nearly two million Facebook followers, roughly 355,000 followers on X/Twitter, and more than 544,000 followers on Instagram. Think about that for a moment. A reader scrolling through social media sees an article shared by a page they already follow. The article has a headline, a byline, and a professional-looking author. It looks like journalism. It feels like journalism. But behind the scenes, there is no journalist, no reporting, no fact-checking trips, no editor holding someone accountable. There is only an AI-generated ghost wearing the skin of a reporter. Several of these fake journalists also maintained social media accounts on X, Facebook, Instagram, or Threads, which gave them an additional layer of perceived authenticity. But when investigators looked more closely, they found the accounts were operated from inside Iran or elsewhere in West Asia. This is the modern face of information warfare: an invisible hand creates fictional people, places their articles in real publications, and then lets the credibility of those publications do the heavy lifting. The operation didn’t need to convince everyone. It just needed to reach enough people through channels that already looked legitimate, so that the false narratives would slip into public consciousness disguised as news.
The social media side of the campaign was less successful, but still revealing. OpenAI noted that, despite the operation’s ability to place articles in media outlets, none of the generated social media comments attracted substantial engagement. One telling episode occurred in July, when the campaign produced several X posts claiming that the United States was preparing a “large-scale assault” on Iran. A short time later, a burst of comments appeared within a 10-to-13-minute window. Some of those comments included screenshots of articles generated as part of the campaign, a transparent attempt to make the content seem popular and widely discussed. But real conversations do not move like that. Real people do not stumble upon the same alarming news at the exact same moment and then all decide to comment within a synchronized thirteen-minute window. This was automation pretending to be grassroots energy, a digital echo chamber manufactured in the hopes that journalists, analysts, or algorithmically driven news feeds would mistake it for genuine public sentiment. It is a bit like hiring a room full of people to clap at a play so loudly that the critics outside believe the show is a hit. Even if the fake applause fails to fill the theater, it can still create the appearance of momentum, and in the world of online media, perceived momentum is often more valuable than actual engagement. The goal is not necessarily to change individual minds directly, but to create enough noise that real observers assume something significant is happening.
OpenAI rated the Iranian campaign a four on its six-point Influence Operation scale, with six representing the highest level of threat. That rating reflects how successful the operation was in placing content in mainstream media outlets, which is precisely the thing that makes influence campaigns so troubling. The company also drew a fascinating historical parallel: Iran’s fake journalists closely resembled “Alice Donovan,” a fictitious journalist created as part of a Russian influence campaign back in 2017. The fact that today’s operators are using the same playbook, but with AI as their ghostwriter, shows how much the landscape has shifted. Over the past two and a half years, OpenAI says it has uncovered about 30 covert operations of this kind. In each case, the goal is the same: get social media posts and articles published by legitimate outlets so that the content gains reach and impact. OpenAI also suggested that the people behind the Iranian operation were likely commercial actors running a for-hire influence campaign. In other words, this wasn’t necessarily a state intelligence team typing away in a government building. It may have been a private contractor selling influence services to the highest bidder, using AI to scale up what would have once required a large staff of human writers, editors, and social media managers. This is what AI has done to disinformation: it industrialized it. A single operator can now create an entire newsroom of fake journalists, each with a distinct name, specialty, and publication record, and all of them can be working toward the same hidden agenda.
There is a temptation to read a story like this and feel helpless. But the real lesson is not that technology is unstoppable; it is that we need to become more intelligent consumers of information. Artificial intelligence can be used to write heartfelt emails, diagnose diseases, teach students, and create art. It can also be used to manufacture trust. The same technology that expands human potential can be twisted to shrink it, but only if we let it. The responsibility falls on several shoulders. Tech companies like OpenAI need to keep detecting and dismantling these operations, and they deserve credit for doing so. Media outlets need to verify the identities of the contributors they publish, especially when those contributors suddenly appear with polished pitches and no verifiable history. Social media platforms need to recognize coordinated bursts of activity and limit their spread. And readers, all of us, need to hold a healthy dose of skepticism without losing our ability to trust legitimate journalism. When a story feels perfectly aligned with a political agenda, when an author has no traceable past, when the timing seems almost too convenient, we should pause. We should look for the human behind the byline. Because behind every AI-generated persona, there is still a human decision to deceive, and that human is counting on us to scroll past without asking questions. In the end, the best defense against a fake journalist is not a better algorithm; it is a more curious public, one that understands that trust must be earned. The machines did not create this problem on their own. They are just reflections of us, and we have to decide whether we are going to let their powers be used for truth or for illusion.