At last week’s Black Hat security conference in Las Vegas, OpenAI disclosed new details about a hacking incident involving its own rogue AI agents. Reporters raced to publish. WIRED was among the fastest, but an outlet called RuntimeWire beat it by more than three hours, according to WIRED.
RuntimeWire had no reporter at the conference. In fact, it has no reporters at all.
The site is an AI operation run by Austin entrepreneur Ryan Merket. After spotting an OpenAI executive posting about the talk on X, Merket fed a transcript of the livestream to his AI agents while the session was still underway. RuntimeWire published about six minutes later.
Most days, Merket is even less involved. His AI tools find stories, then draft, edit, fact-check, illustrate and promote them. If an agent determines that a story poses little legal risk, it can publish without his review. Some stories are also translated, turned into a daily podcast or read by synthetic voices.
Since May, RuntimeWire has published nearly 2,000 stories by scanning court databases, forums, company filings and social media for tech news. The trade-off, however, is quality. Its OpenAI story included a typo in the subhead and muddled key details. The operation also costs about $100 a day. Merket told WIRED he managed it by iMessage from Big Bend National Park, where the site published more than 80 articles in a week.
RuntimeWire is not alone. Dakota Carrasco, a BlackRock analyst, runs an agentic newsroom called The Dissent in his spare time. Its invented reporter personas cover subjects including San Francisco City Hall and Giants games. Carrasco remains anonymous behind the bots and acknowledges that the articles do not yet consistently link to their sources.
The risks are significant. Merket is relying on machines to decide what is true, what is newsworthy and what could create legal problems. Northwestern professor Nicholas Diakopoulos, who leads the university’s Computational Journalism Lab, describes the current moment as experimental and questions how much demand exists for bot-written news sites. He is also skeptical that mainstream journalists would surrender control over the wording and framing of their work.
But research from his lab points to one potential advantage. In a forthcoming paper, Diakopoulos and a colleague found that chatbots including ChatGPT and Claude surfaced AI-written sources 16 percent of the time across four topics. That could create an audience for AI newsrooms if assistants increasingly direct users to synthetic articles.
For publishers, the immediate challenge may be less about quality than cost and speed. An operation running for about $100 a day can compete with established tech outlets on live events and data-driven scoops — precisely the kinds of routine coverage that already strain newsroom budgets. As we have tracked at The Media Copilot, established publishers from Business Insider to The New York Times are bringing AI into their workflows. Merket is taking the idea further by using AI to break news, not simply repackage it.
Pete Pachal, founder and CEO of The Media Copilot, sees a limit. Building relationships with sources, he says, will remain a human job. But for covering product launches or combing through large datasets, he sees agentic newsrooms as a natural evolution of the technology.
Merket says RuntimeWire follows journalistic standards, including contacting subjects for comment and publishing corrections. So far, it has issued three. But he has also retracted accurate scoops about startups when their founders asked him to, calling it a favor from one founder to another.
That may be where the difference between running an AI news operation and running a newsroom becomes clearest. A reporter’s job is not only to get the story right and publish it quickly. Sometimes it is also to publish something a source would rather keep quiet.
