The Black Hat security conference room was full of reporters from major outlets taking notes while OpenAI walked through the details of the Hugging Face breach live on stage. None of them got the scoop. It went to RuntimeWire, an AI news site whose newsroom doesn't employ a single journalist, because the newsroom is a piece of software. Kate Knibbs tells the story in Wired.
A newsroom that costs $100 a day
RuntimeWire was launched in May by Ryan Merket, an Austin entrepreneur and investor with stints at Reddit, AWS and Microsoft, and he remains the only human on staff. Everything else is a chain of AI agents that monitors around fifty sources (corporate blogs, changelogs, research papers, social feeds), filters, scores every story against editorial criteria, digs into primary sources, writes, edits and fact-checks. 'The newsroom itself is software,' Merket sums up at The Media Copilot, 'a pipeline of AI systems that scans sources, researches, writes, edits and fact-checks around the clock.'
The experiment's numbers: 71,796 stories scanned since May, 1,627 published (2.3%, so the machine rejects too), around 205,000 pageviews and a running cost of roughly $100 a day, per Wired. Revenue so far: zero, though Merket is already testing sponsorships and a jobs board. A good chunk of the output goes live without any human reading it first.
The scoop: a QR code, 6 minutes and a 3-hour head start
The play that put RuntimeWire on the map is this. During OpenAI's Black Hat session, Merket spotted someone on X sharing a QR code linking to the talk's live caption stream. He grabbed the transcript, fed it to his system, and the agents researched, drafted, edited and fact-checked the piece in about 6 minutes. By the time Wired published its own story, the machine's version had been sitting on Google for over 3 hours.
'I was moving really fast because I knew there were reporters in the audience trying to scoop it as well,' Merket admits to Talking Biz News. And note the detail, every article carries his byline, even though his synthetic team does the writing.
The machine types fast, but the nose for news is still human
Those who read the piece closely, like Gizmodo, describe it as flat, mechanical prose, a play-by-play with no reading between the lines. And there's one detail I'd call the heart of this story, the AI didn't smell the scoop. Merket did, the man who saw the QR code on X and instantly understood there was a story in it. The machine just typed very fast what a human had already hunted down.
I say this from experience, because we use AI daily around here and don't hide it, the tool speeds up everything that comes after the judgement, but the judgement doesn't automate. Choosing what deserves telling, from what angle and with what opinion, is still people's work, just like deciding what gets labelled as AI-made or how generated text gets watermarked still is.
The conversation starts here
Sign in with a supporter account to comment. Sign in
Nobody has commented yet. Want to go first?