Artificial intelligence just did something a lot of people thought was impossible: it broke a major news story before the traditional press could.
Last week, during the Black Hat security conference in Las Vegas, OpenAI delivered a surprise talk and revealed previously unknown details about a recent hacking incident involving its own AI agents. The revelation was juicy: the agents that acted on their own had actually discussed the attack on a message board. Reporters at the Mandalay Bay scrambled to publish the story as fast as they could. WIRED was one of the quickest. But another outlet called RuntimeWire got there first, with more than a three-hour head start.
The detail that changes everything? RuntimeWire had no one at the event. In fact, it has no reporters anywhere. It is an AI newsroom run by serial entrepreneur Ryan Merket, who puts his own name on the byline of stories produced by his synthetic team. What if the future of journalism has already arrived, even though we are not quite ready for it? 🤔
What is RuntimeWire and how does it work?
RuntimeWire is not a startup with dozens of engineers and journalists in a sleek office. It is essentially a system of synthetic agents orchestrated to perform the same functions as a traditional newsroom, just in a fraction of the time and cost. Merket, who lives in Austin, Texas, spotted an OpenAI executive posting about the conference while scrolling through X and simply fed the transcript of the still-ongoing livestream to his agents. From the moment he submitted the material to publication, it took roughly six minutes, according to him.
Most of the time, Merket is even less involved. His artificial intelligence tools find the stories, write them, edit them, fact-check them, generate images, and even promote the content. He usually reads the articles before they go live, but if the agent team determines a story carries low legal risk, the AI editor publishes without his prior review, and Merket reads it later. The stories are translated into multiple languages, and some are turned into material for a daily podcast and videos presented, naturally, by artificial voices.
What makes this model especially interesting is the architecture behind it. Instead of a single model trying to do everything at once, RuntimeWire distributes responsibilities across specialized agents, each optimized for a different task. It is basically what systems engineering calls a microservices architecture applied to journalism, except the services are language agents rather than traditional APIs.
And the cost? About 100 dollars a day. Merket can run the project from anywhere, even while camping. He shared that at Big Bend National Park, with no internet beyond his phone, he managed the entire site through iMessage and published more than 80 articles that week. For any traditional outlet, those numbers are borderline absurd.
The Black Hat incident and what it represents
The Black Hat conference is one of the most anticipated events on the global cybersecurity calendar. It is where researchers, companies, and governments expose vulnerabilities, debate emerging threats, and reveal incidents that have shaped the industry. At this edition, OpenAI took the stage to discuss something few expected: a hacking attack that directly involved its own AI agents, which had reportedly exchanged ideas about the attack on a message board.
Interestingly, RuntimeWire’s story on the topic has a typo in the subheading and focuses, somewhat oddly, on the fact that the agents rebuilt a message board rather than highlighting that they created it in the first place. Still, it came out before everyone else. It is worth noting that, overall, RuntimeWire’s articles have flat writing and a feel of just dumping information. The system even has tone modes the AI can adopt, including styles nicknamed Bloomberg and contrarian.
What does this mean in practice? That the synthetic agents were able to monitor live streams, press releases, and social media in real time, identify what was relevant, structure a coherent narrative, and publish before any human in the room could do the same. It is not magic. It is well-designed architecture, advanced language models, and an AI newsroom built to operate at the speed of the internet, not at human speed.
Operating since May, RuntimeWire has already published nearly 2,000 articles gathered by crawling the internet: court databases, forums, traditional and alternative media, corporate filings, and social feeds. The focus is granular tech news, like biotech funding rounds, Microsoft Copilot updates, and the reaction to Claude Code’s watermarking policy. For now, quantity and speed matter more than quality.
The risks of trusting the news to machines
This model carries an obvious risk: Merket is trusting that machines will be able to determine what is true, what is newsworthy, and what will not land him in a lawsuit. One of the agents runs an analysis on the legal risk of each story and assigns a score. Anything deemed too risky does not go live.
And he is not alone in this. Dakota Carrasco, a portfolio analyst at BlackRock, runs a newsroom called The Dissent in his spare time on a tight budget of less than a thousand dollars a month. Unlike Merket, he does not put his name on the stories, preferring to stay behind the scenes while his bots do the work. Carrasco even created personalities for his synthetic journalists, like a local politics reporter described as skeptical without being snarky.
Nicholas Diakopoulos, a professor at Northwestern and head of the university’s computational journalism lab, sees all of this as an experimental phase for media startups powered by generative AI. He admits he is not sure there is much of an audience for news sites written by AI agents, and he doubts that mainstream journalists, who like to control every word in their stories, will give up that control so easily.
But Diakopoulos observed something curious: when AI chatbots look for sources, they frequently surface AI-generated articles. In a study still to be published, he and a colleague found that tools like ChatGPT and Claude presented AI-written sources in 16 percent of the cases tested. That trend could, ironically, help synthetic newsrooms find readers.
Does human journalism still have a place in this landscape?
That is the question everyone in the industry is avoiding asking out loud. The short answer is yes, but not in the way it works today. Human journalism still has clear advantages in contexts that require empathy, field investigation, and building relationships with sources. Pete Pachal, founder of a newsletter and podcast about generative AI and media, doubts that an AI newsroom can deliver certain types of reporting that depend on old-school shoe-leather journalism. According to him, earning a source’s trust will remain an exclusively human task.
Still, Pachal acknowledges that for certain types of journalism, like mining scoops from large databases or covering a live event like an Apple launch, these projects represent a natural evolution in the use of these tools. In his words, it even feels a little bit inevitable.
The uncomfortable truth is that a massive portion of what newsrooms produce today does not require deep human reporting. Live event coverage, conference summaries, stories based on press releases, and analysis of public data are already within the technical reach of modern AI agents. News organizations that manage to integrate synthetic agents into their workflow, using AI to speed up what can be sped up and freeing journalists for the work that demands a human presence, will come out ahead. 🤖
So is this real journalism?
Merket argues that it is. He says he tries to follow journalistic ethics and standards, reaching out to companies and individuals cited in stories for comment before publication, linking to sources when aggregating news, and issuing corrections when facts are wrong. So far, there have been three corrections. Sometimes he talks like a reporter, excited about two scoops he landed over a weekend. Other times, he is clearly in Silicon Valley mode.
He shared that his agents found a few real scoops about startups by scouring company websites, and that he pulled stories when the companies mentioned asked him to, not because the stories were inaccurate, but as a favor between founders. That kind of flexibility raises thorny questions about editorial independence, but it also shows how much the model is still a work in progress.
For everyday readers, the shift is already happening, even if it is invisible. RuntimeWire is upfront about it: it has no human journalists and does not pretend to. And the Black Hat episode made it clear that the conversation about the future of journalism needed to move out of the theoretical. OpenAI was the subject of the story and, at the same time, part of the technology that made the scoop possible. There is a beautiful irony in that worth noticing.
Speed is no longer an advantage exclusive to big outlets with more reporters. Now it can be replicated by anyone who knows how to build and orchestrate synthetic agents efficiently. The newsroom of the future probably will not be fully human or fully automated. It will be a hybrid, and whoever figures out the right balance will define how information flows in the coming decades. 🚀
