Iranian operatives used AI to plant fake op-eds in U.S. media outlets

A Florida newspaper published an opinion piece that read like the work of a concerned American citizen. It wasn't. Iranian operatives had used artificial intelligence to generate the text, then placed it in the outlet as part of a coordinated campaign to shape American public discourse without detection.

This wasn't an isolated incident. According to reporting from OpenAI and security researchers, Iranian state actors have systematically leveraged ChatGPT and similar AI tools to manufacture fake op-eds and plant them across multiple U.S. media outlets. The operation represents a significant escalation in how state-sponsored disinformation works in the age of generative AI—no longer requiring human writers, native speakers, or the kind of obvious tells that once made foreign propaganda recognizable.

The mechanics are straightforward and efficient. An operative writes a prompt in ChatGPT describing the narrative they want to push, the outlet they're targeting, and the tone they need. The AI generates plausible text in fluent English. The operative submits it to an American newspaper under a fabricated byline, often posing as a local resident or expert. If the outlet's editorial staff doesn't catch it—and many don't, given the volume of submissions they receive—the piece runs. Readers see what appears to be legitimate opinion journalism. The influence spreads.

What makes this campaign particularly concerning is its scale and sophistication. Russian actors have also been caught using ChatGPT for similar influence operations, suggesting this is becoming a standard tool in the state-sponsored disinformation toolkit. The targeting of multiple outlets means the operation wasn't testing a concept; it was executing a strategy. Each successful placement amplified the reach and credibility of narratives that served Iranian interests, whether on foreign policy, sanctions, or American domestic divisions.

The challenge for media outlets is acute. Editors receive hundreds of submissions weekly. An AI-generated op-ed that's well-written, topical, and submitted under a plausible name can pass initial screening. The text itself contains no obvious errors or tells. It doesn't read like a translation. It doesn't sound robotic. It sounds like opinion journalism, because it was generated by a system trained on millions of examples of opinion journalism.

OpenAI's detection of these campaigns came through monitoring for unusual patterns in how accounts were being used—multiple submissions from different accounts with similar characteristics, submissions that didn't match the stated identity of the author, geographic inconsistencies. But that kind of forensic analysis happens after publication, if it happens at all. Most newspapers don't have the resources or expertise to conduct it.

The broader implication is that the barrier to entry for state-sponsored influence operations has collapsed. A decade ago, running a disinformation campaign required hiring writers, managing networks of fake accounts, and hoping the work was good enough to fool editors. Now it requires an internet connection and a willingness to violate OpenAI's terms of service. The cost is near zero. The potential reach is enormous. And as AI systems improve, the detection problem only gets harder.

Media outlets and technology platforms are beginning to grapple with how to respond. Better verification of author identity, more rigorous fact-checking of submitted opinion pieces, and closer coordination with platforms like OpenAI about detecting coordinated inauthentic behavior are all being discussed. But the fundamental problem remains: AI-generated text is becoming indistinguishable from human-written text, and the incentives for state actors to exploit that gap are only growing stronger.