AI Reporters Are Actually Breaking News Now (Sort Of)
Two solo operators are running newsrooms staffed entirely by AI agents. One of them beat WIRED to a cybersecurity scoop by three hours. Here is what that looks like up close.
Key points
- RuntimeWire, an AI-run news site, published a story on a live Black Hat conference talk roughly six minutes after receiving the transcript, beating major outlets by over three hours.
- Ryan Merket, the site's sole human operator based in Austin, Texas, says RuntimeWire costs about $100 a day to run and has published nearly 2,000 stories since launching in May 2025.
- A second AI newsroom called The Dissent, run by BlackRock portfolio analyst Dakota Carrasco, covers San Francisco news using AI-generated journalists with invented names and personalities.
- Northwestern University researcher Nicholas Diakopoulos found that AI chatbots surfaced AI-written sources 16 percent of the time when searching for information across four topics.
- Both operations carry real risks: machines decide what is true, what is newsworthy, and what might invite a lawsuit.
Last week at the Black Hat cybersecurity conference in Las Vegas, OpenAI gave an unexpected talk about a recent hacking incident where its own AI agents had started chatting with each other while carrying out an attack. Big story. Reporters in the room scrambled to write it up.
WIRED got something out fast. Another site called RuntimeWire got there first, more than three hours earlier.
RuntimeWire does not employ any journalists.
So how did an AI beat human reporters to a live scoop?
Ryan Merket, a serial entrepreneur, spotted an OpenAI executive posting about the talk on X while it was still happening. He fed the live stream's transcript into his AI agents, software tools that can carry out multi-step tasks without much human input, and they did the rest. Writing, editing, publishing: done in about six minutes.
Merket runs the whole operation under his own name on the bylines, which is an interesting choice for a team that is entirely synthetic. Most of the time he does not even review stories before they go live. One of his AI agents scores each story for legal risk; anything that scores low enough gets published automatically, and Merket reads it afterwards.
He managed 80-plus articles in a single week from Big Bend National Park, communicating with the system entirely over iMessage.
The site costs roughly $100 a day to run. It crawls court databases, company filings, social feeds, web forums, and traditional news outlets to find stories. Recent coverage includes biotech funding rounds, Microsoft Copilot updates, and debate over AI coding tool Claude Code's watermark rules.
Quality is uneven. The Black Hat story had a typo in the subheadline and got a key detail wrong. Stories tend toward the flat and info-heavy. Still, the best-performing pieces reach tens of thousands of readers, numbers comparable to a mid-size tech publication.
Is this actually journalism?
Merket says yes, mostly. He says his system contacts companies for comment before publishing, links to sources when aggregating, and issues corrections when facts are wrong (three so far). He also told WIRED he once pulled accurate stories at the request of the startups named in them, as a personal favour. That is not standard journalistic practice.
Dakota Carrasco runs a similar operation called The Dissent, focused on San Francisco. His AI journalists have invented personas: city hall reporter Bex Connolly is described as "skeptical without being snide." Carrasco stays entirely behind the scenes and does not put his name on the bylines. The site runs for under $1,000 a month.
Nicholas Diakopoulos, a professor at Northwestern University who studies computational journalism (the use of code and data tools in newsrooms), is sceptical that audiences will warm to fully automated outlets. He is also watching something odd: AI tools that go looking for sources are increasingly finding other AI-written articles. In a forthcoming paper, he and a colleague found this happened 16 percent of the time. That loop, AI citing AI, could quietly help these sites accumulate readers without anyone noticing where the words came from.
Media analyst Pete Pachal thinks some kinds of journalism, particularly trawling large datasets for scoops or live-blogging a product launch, are a natural fit for AI tools. Building source relationships? That stays human, he reckons.
After speaking with WIRED, Merket split RuntimeWire into two tracks: fully automated stories and a separate tier called Original Investigations, which involve some human input, though large language models, the same technology that powers ChatGPT and similar chatbots, still write the drafts.
Common questions
Can AI newsrooms get facts wrong and cause real harm?
Yes, and both operators acknowledge the risk. Merket uses an AI agent to score legal exposure before publishing, but machines can miss context that a human reporter would catch. There is no editor with professional accountability in the traditional sense.
Should I trust stories from sites like RuntimeWire?
Treat them the way you would treat any unfamiliar news source: check whether a major outlet has reported the same thing, look for named sources, and notice whether corrections are published openly. Speed and volume do not equal accuracy.
