There is a strange paradox at the heart of the artificial intelligence moment: the same technology that can help a student draft an essay, a doctor spot a disease, or a small business translate its webpage into a dozen languages is also being used, quietly and systematically, to manipulate public opinion, spread confusion, and tear at the fabric of democratic debate. Recent disclosures by OpenAI have pulled back the curtain on two covert influence campaigns, one traceable to Russia and another to Iran, both of which used AI-powered chatbots to manufacture fake personas, pitch articles, write comments, and poke at some of the most sensitive geopolitical fault lines in the world. The picture they paint is not one of superintelligent machines taking overt the globe, but something perhaps more unnerving: ordinary, determined propagandists using AI as a force multiplier, churning out identity fragments and misleading material that can slip, sometimes with startling ease, into mainstream publications and social media feeds. This is the context in which a “Must Read” headline has been making the rounds, declaring that President Trump has described people who use the term “Artificial Intelligence” as “THE ENEMY,” and asking what his new AI order actually means. Whatever the precise policy calculus behind such a phrase, it speaks to a growing cultural anxiety about AI, an unease that is entirely justified and yet also dangerously vague. If we treat the term itself as an enemy, we risk missing the actual enemies: targeted influence networks, fabricated authority, and platforms that allow malicious content to travel further and faster than ever before. Another “Must Read” item in the same moment reminds us that Anthropic, the maker of Claude AI, astightened its own rules on model abuse, election meddling, weapons, and surveillance, suggesting that AI developers increasingly see themselves asthe new border guards of the information landscape.
The first operation disclosed by OpenAI, named “Dark Clark,” involved a cluster of Russia-origin ChatGPT accounts that the company said it had banned. The activity was aimed primarily at countries across Latin America, with much of the content apparently designed to undermine Ukraine’s reputation and to reset geopolitical narratives in ways favorable to Moscow. There were also efforts focused on domestic politics in Argentina and Bolivia, where staff could sow division by amplifying local grievances, by pretending to be part of conversations they were never meant to be part of, and by generating a slow drip of commentary that looked, at first glance, like organic public sentiment. To make this web of fake accounts feel real, the operators appeared to control a purported research platform called the Social Research Center, or SRC, through a fabricated persona named”Mia Clark.” This invented academic leader gave the operation a sheen of legitimacy, an institutional address, and a human face. The platform’s stated interests at first sounded almost boringly benign, focusing on the Indian diaspora in Latin America and on relations between India and countries in the region, a topic that would attract little scrutiny while allowing the operators to move within policy circles, scholarly chats, and community forums without standing out. But the true intent was likely much grimmer: use a neutral-sounding research body to gain access, to build a reputation over time, and to slowly inject material that served Russian strategic interests. What is especially troubling, OpenAI said, was that the available evidence suggested that employees working on the ground in Latin America may have been unaware of the Russian operation entirely. They may have believed they were working fora legitimate research institute, participating in surveys, translating documents, or moderating discussions, when in reality they were pawns in an influence game orchestrated thousands of miles away. That human dimension turns the story from a purely technical curiousity into something visceral: how easily good-faith people can be drawn into bad-faith projects when the infrastructure of trust has been quietly hollowed out.
The “Dark Clark” campaign also made heavy use of fake documents, fabricated audio scripts, and other misleading material, all designed to support narrative lines in the region. Among the most specific targets was an ongoing stream of content related to claims about Ukraine’s honorary consul in Panama, a subject that at first seems obscure but that actually carried symbolic weight in Latin American media ecosystems, where Panama’s diplomatic posture, Russian influence, and Ukrainian solidarity intersect in complicated ways. The material was not necessarily sophisticated by the standards of a spy novel, but it did not need to be. In an online environment already saturated with conflicting claims, a plausible fake document or a well-timed audio clip can germinate like a weed, especially when shared by accounts with established personas. OpenAI rated this operation Category 5 on its six-point influence-operation scale, which deserves pause: that is the highest rating the company has assigned to any campaign since it began reporting on such activity. It does not necessarily mean the campaign was stunningly effective in changing minds; rather, it appears to reflect the brazenness, the operational depth, and the persistence of the effort, as well as the relative difficulty of unraveling its layers. One of the uncomfortable lessons here is that influence operations no longer require a Kremlin newsroom with a thousand employees and state television crews. A small team, or even an individual, can use language models to generate plausible news articles, academic bios, comment sections, and social media posts in multiple languages, each tailored to its audience, while using AI to lower the cost of scale. The enemy, in other words, has become bureaucratic and modular, and it operates under so many layers of invented identity that even the people feeding it information might not know where it ultimately leads
The second operation, named”Bogus Bylines,” carries an almost mordantly perfect title, because it was built on the oldest currency in journalism:a byline, a named author, a voice that the reader trusts until given a reason not to. OpenAI said it had banned Iran-origin accounts linked to this campaign, which used seven fabricated journalist identities to pitch geopolitical articles, largely about the US-Iran conflict, to online publications. These fake journalists had names, presumably profile photos, possibly fabricated credentials; they wrote in a plausible, newsy tone; they approached editors with what looked like an article already ready to publish, saving cash-strapped outlets the time and effort of assigning a real reporter. The gambit worked far beyond what anyone should be comfortable with: OpenAI identified almost 100 articles that were published or syndicated under the operation’s bylines across roughly a dozen outlets. That means real websites, with real audiences, ran pieces by nonexistent journalists, often about matters of international conflict, without apparently catching the deception. The operators also generated batches of political comments beneath these articles, presumably trying to manufacture the appearance of reader engagement, debate, and outrage, the classic tactic of astroturfing a reaction to make a story seem more important than it actually was. But here is a small mercy: OpenAI found little evidence that these AI-generated comments attracted substantial engagement from actual readers. The bots were, in a sense, talking to themselves, a few lonely voices in an empty digital room. Yet the operation still matters, because even if public interaction was weak, the publication of those articles gave hostile actors a legitimate-looking archive, an appearance of journalistic credibility, and a persistent online footprint that could be cited, shared, and weaponized long after the original publication dates. Perhaps the most human detail in this entire report is that internal reports from the operation appeared to inflate the campaigns’ effectiveness through misleading calculations. In other words, the propagandists were lying to their own superiors about how well they were doing, cherry-picking engagement metrics and presenting vanity numbers as proof of success. It is almost comic, but it jolts us back to reality: these operations are run by human beings with organizational pressures, career anxieties, and incentives to exaggerate, just like every other institution. AI did not create those impulses; it merely gave them a faster, more scalable way to perform failure as success, while making it harder for outsiders, and insiders alike, to know what actually happened
What should we take away from these revelations, aside from a queasy sense of having watched a magic trick and then being shown the wires? The first lesson is technological: AI is making covert campaigns dramatically easier to organize. Translation, localization, persona generation, writing at scale, and even the creation of fake audio and documents, all of which once required significant human labor, can now be done by a handful of operators with a subscription to an API. The second lesson is institutional: established publications can give misleading content a far wider potential reach than any bot network could achieve on its own. By running articles bylines from fake journalists, outlets unknowingly lent their reputations to foreign influence efforts. That is not an indictment of every editor who was fooled; newsrooms are stretched, wire services are trusted, and the web has trained us to value speed over verification. But it is a reminder that media literacy cannot stop at the reader; it must extend to editors, publishers, and platforms, who need to ask harder questions about who is really behind a pitch, a name, and a source. This is also why companies such as Anthropic are tightening their policies, with Claude AI’s new rules on model abuse, elections, weapons, and surveillance, acknowledging that the same models that write cheerful poems can also draft targeted political ads, identify vulnerable people, or inadvertently assist weapons development. These rules will not stop every bad actor, but they represent a necessary effort to place guardrails around powerful technology before the damage becomes chronic. Meanwhile, the “Must Read” question about Trump’s new AI order remains hanging over all of this: what does it mean when a leader calls the phrase”Artificial Intelligence” the enemy, rather than distinguishing between the tool and the misuse? If the answer is simply suspicion of anything labelled AI, we may end up stifling innovation, handicapping trustworthy developers, and leaving the field open to exactly the adversarial actors these new policies are meant to catch. If, on the other hand, the point is that we should stop treating AI as a magic buzzword and start treating it as a mundane, fallible, dangerous-but-usable tool, then perhaps the phrase is less absurd than it sounds.
At its core, this story is not about algorithms at all. It is about trust, about the slow, fragile process by which human beings decide what to believe, and about how easily that process can be gamed when the machinery of authorship is cheap and the machinery of verification is expensive. The Russian operators behind Dark Clark and the Iranian operators behind Bogus Bylines were not geniuses of deception; they were opportunists who recognized that an invented research center, a fake byline, and a plausible-sounding paragraph could pass through an information ecosystem that has become to crowded to inspect every passing voice. The real defense, then, is not to ban the term AI, nor to assume that every AI-adjacent tool is inherently sinister, but to rebuild habits of verification, humility, and patience in our public discourse. Readers can learn to ask where a piece of information came from, who paid for it, what assumptions it rests on, and whether the person sharing it has a real biography. Editors can build stronger provenance checks, especially for wire copies, opinion pieces, and international correspondents. AI companies can continue to tear down networks, share threat intelligence, and, as OpenAI said it had done, turn over findings to relevant authorities, helping law enforcement and researchers map these hidden webs. And governments, regardless of their slogans about AI, can focus on deterrence and resilience rather than either naive embrace or reflexive paranoia. The phrase “artificial intelligence” may invite fear, but the real question has never been whether machines are our enemies. It is whether we, as a society, are willing to do the unglamorous, human work of paying attention, of refusing to let fabricated voices speak evenly with real ones, and of remembering that behind every effective propaganda operation, there are people who believe that convenience matters more than truth. AI has made it easier for them to act on that belief. It has also made it easier for us to expose them, if we choose to look. The story of Dark Clark and Bogus Bylines is ultimately a story of people, not programs, and it ends, as all such stories do, with a choice about what kind of information world we want to build, one where fabricated identities can quietly shape politics, or one where trust is earned, verified, and protected as the essential infrastructure it has always been.