Imagine a small-town newspaper editor in Florida, someone who has spent years building relationships with readers, contributors, and community members. He receives an opinion piece through a familiar organization that has supplied his paper with thoughtful content for a long time. The essay is well-written, measured, and seems to come from a concerned citizen using a name that sounds perfectly real. He does not check the author’s identity, because he has no reason to. The organization is trusted. The prose is compelling. The topic is important. But the byline belongs to a person who never existed, and the words were generated by artificial intelligence, shaped and polished by unknown operators working on behalf of a foreign influence campaign. This is not a hypothetical scenario dreamed up by tech ethicists. It is exactly what happened recently when an Iranian-linked operation used ChatGPT to produce about one hundred opinion pieces under invented identities. The articles were placed in more than twenty outlets, from a tiny weekly newspaper serving fifteen thousand people to larger online platforms. The operation did not need to hack into a database or steal secrets. It simply exploited something far more fragile and far more precious: the human habit of trust.
The campaign was careful and strategic. According to OpenAI, the company behind ChatGPT, the operation was linked to users in Iran and focused heavily on the war with Iran and its possible costs for Americans. The articles were not loud propaganda; many were framed as anti-war commentary, designed to seem reasonable, reflective, and even sympathetic to American readers. The people behind the operation did not stop at drafting text. They used ChatGPT to refine the articles, adapt them to the specific requirements of different publications, and draft pitch messages to editors, mimicking the polite, professional tone that real contributors use when submitting work. One of the fabricated bylines was Ervin Hoskins, whose name appeared on an opinion piece in the Port St. Joe Star, a weekly paper in Gulf County, Florida. The publisher, David Adlerstein, later told the Washington Post that the submission came through PeaceVoice, a liberal Oregon-based organization that had supplied material to his paper for years. He did not check the article’s source or confirm the author’s identity, because the group’s longstanding relationship with the publication had earned his trust. The piece warned that the costs of war would continue to affect Americans long after the fighting ended. It sounded exactly like the kind of thoughtful op-ed a local paper would want to publish.
The operation did not stop with one small Florida weekly. OpenAI said Hoskins was one of seven names connected to the campaign, and the fabricated content reached larger online publications as well. Middle East Monitor, a website that describes itself as supportive of the Palestinian cause, published articles attributed to false authors before taking them down after being alerted. Daily Kos, a major progressive community blog, carried some of the material through a community platform that does not undergo pre-publication review. Markos Moulitsas, the founder of Daily Kos, said the operation undermined trust on the platform and encouraged users to suspect one another of being bots whenever they disagreed. That loss of trust may be the most dangerous consequence of all. In a healthy public conversation, people can argue passionately and still believe they are arguing with a real human being. But when AI-generated identities enter the mix, every disagreement becomes suspect. The reader cannot tell whether the person on the other side of a debate is a concerned citizen or a language model. The editor cannot tell whether the byline belongs to a retired professor or a chatbot. The operation targeted the editorial process, but its deepest effect is on the shared sense of reality that makes democratic debate possible. For small outlets, the harm is especially severe. They have the same responsibility to be credible as national newspapers, but they do not have the same resources to verify every submission. The warning signs are often invisible, because modern AI can imitate human cadence, express opinions, and make arguments in ways that are nearly indistinguishable from authentic writing.
OpenAI’s response was quick but necessarily limited. The company banned the accounts linked to the activity, restricted access to ChatGPT from Iran, and shared the information with relevant authorities. At the same time, it acknowledged a troubling gap: it could not determine who was ultimately responsible. The operation may have originated in Iran, but the exact actors—state agencies, independent operatives, or something in between—remain unknown. OpenAI said the purpose of publishing its findings was not to claim a victory but to alert regulators, businesses, and the public to the ways hostile actors are exploiting generative AI. The company also pointed to weaknesses in media workflows that allow fabricated material to pass as genuine. That is a remarkable admission from the maker of the very tool used in the campaign. It is also a reminder that the defense cannot be left to any single company or government. The technology is cheap, accessible, and constantly improving. Generating a hundred plausible articles costs almost nothing and requires no special technical skill. Verifying a hundred submissions, by contrast, is expensive, time-consuming, and impossible to do perfectly. This asymmetry is the heart of the new threat. Offensive operations are scalable and fast; defensive measures are slow and labor-intensive. OpenAI’s actions may stop this particular campaign, but they cannot stop all campaigns, and they cannot restore the trust lost when readers discover that a publication they respected has published a fake.
The broader lesson is not limited to one country, one platform, or one political cause. As AI tools become more capable and more accessible, the burden on editors to verify names, sources, and submission histories has become far heavier, even in small newsrooms with limited resources. The traditional assumption that a submission from a known organization or a polished pitch is probably legitimate is no longer safe. But the answer is not to close the gates entirely, because independent and local media depend on contributed content, especially as newsroom budgets shrink. Instead, this incident suggests the need for a new kind of literacy, both inside news organizations and among readers. Editors should ask harder questions. Does this author have a verifiable track record? Has the same name appeared in other publications? Does the submission include any signs of automation or unusual requests? Some of these checks can be automated, but many require human judgment. Publishers also need to be transparent with their audiences about their policies on artificial intelligence and contributed content. In a small paper, that might mean publishing a simple note: we verify the identity of all first-time contributors. On a large platform, it might mean reviewing posts before they go live. The goal is not to make publishing impossible, but to restore a degree of accountability. At the same time, technology companies need to build detection tools and watermarking into their products so that AI-generated material can be identified after the fact. Media associations can help by sharing information about known influence operations. But all these measures take time, money, and coordination, while the attackers need only a few hours and an internet connection.
Still, there is something deeply human about this story, and that is precisely what makes it unsettling. The Iranian-linked operation did not use complicated espionage or violent coercion. It used the same techniques that have always worked in politics: create a plausible messenger, say what people want to hear, and rely on goodwill to get through the door. The anti-war framing was not an accident; it was a calculated choice to make the fake articles seem more credible and more like authentic public debate. The fact that a small Florida paper and a large progressive blog both carried the content shows that the operation understood its audience. It also shows that AI is not a magic wand but a force multiplier for a very old kind of deception. The danger is not that readers will immediately believe every AI-generated argument. The danger is that they will stop believing any of it. When trust collapses, the public square becomes a collection of isolated individuals shouting past each other, unable to agree even on the existence of a common reality. That is the real prize for a hostile actor, and it is far more valuable than a handful of op-eds. The response must therefore be as much about protecting relationships as about detecting fake text. Editors need to know their contributors; readers need to know their sources; platforms need to know their users. This will not be easy, and there will be more incidents like this one. But the story of Ervin Hoskins, a man who never existed, reminds us that the most important fact in any article is not always the first line. It is the name at the top and the trust that name carries. As AI grows more fluent, that trust will have to be earned again and again, not just with better algorithms but with the old-fashioned virtues of care, verification, and honesty. The future of informed citizenship may depend on it.