AI Newsroom Scoops Human Reporters at Security Conference

At the Black Hat security conference in Las Vegas last week, OpenAI disclosed details of a recent hacking incident in which its rogue AI agents had discussed their attack on a message board. The news broke in real time, and reporters in the audience scrambled to file stories. WIRED was among the first, but RuntimeWire—an AI-operated newsroom with no human journalists on staff—published first, beating traditional outlets by more than three hours.

RuntimeWire is run by serial entrepreneur Ryan Merket, who puts his name on bylines written entirely by AI agents. Merket spotted an OpenAI executive posting about the conference while scrolling X, then fed the live transcript to his agents while the talk was still happening. The story published in "about six minutes" with no human prepublication review—the AI editor determined it posed minimal legal risk and published autonomously. (The story itself contained a typo in the subhead and oddly emphasized that agents rebuilt a message board rather than created one.)

Since launching in May, RuntimeWire has published nearly 2,000 stories by crawling the internet—court databases, web forums, social feeds, company filings, and more—all sourced, drafted, edited, fact-checked, and published by AI with no human reporters involved. Merket typically reads stories after publication, but when his AI editor deems legal risk low, the story goes live without his review. Stories are translated into multiple languages and repurposed into a daily podcast and videos narrated by artificial voices. The operation costs about $100 per day to run. Merket managed it remotely from Big Bend National Park using only his phone via iMessage, publishing over 80 articles that week.

The quality is uneven. Stories are flatly written with an "info-dump" quality, and the backend supports multiple tonal modes ("Bloomberg," "contrarian," and others) that the AI can adopt. Yet the hits work: some stories attract audiences comparable to midsize tech websites, with tens of thousands of readers, while duds generate little traffic. Merket previously worked on ads at Reddit and is building audience through tech-themed Subreddits. This week, after speaking with WIRED, he split the newsroom into two tiers: fully automated news and "Original Investigations" that involve some human reporting but are still drafted by large language models.

Merket is not alone. Dakota Carrasco, a BlackRock portfolio analyst, runs an "agentic newsroom" called The Dissent in his spare time, focused on San Francisco news. Launched in March, The Dissent costs under $1,000 per month and employs synthetic journalists with distinct personas: beat reporter "Bex Connolly" (skeptical without snide) and "Sal Moreno" (sports degenerate delivering Giants news with no "bro-science" or "Rogan-style credulity"). Unlike Merket, Carrasco does not publish under his own name. The Dissent emphasizes aggregation but without strong citation norms—bots mention sources but rarely hyperlink.

Experts are divided on whether this works. Northwestern professor Nicholas Diakopoulos, who runs the Computational Journalism Lab, calls it an "experimental phase" but doubts mainstream journalists would cede control to AI agents, and he's skeptical of audience demand for AI-written news. He has also found a troubling pattern: when AI tools like ChatGPT and Claude search for sources, they surface AI-generated articles 16 percent of the time across different topics, potentially feeding synthetic content into AI newsrooms in a self-reinforcing loop. Pete Pachal, who runs a newsletter and podcast about generative AI and media, sees potential in narrow domains—live-blogging product launches or mining large datasets—but doubts AI agents can handle reporting requiring human trust and sourcing. "Cultivating the trust of a source, I do think that's going to be human-only," he says. Yet he calls these projects a "natural evolution" and says it "feels a bit inevitable."

Merket insists he is following journalistic ethics: he contacts sources for comment, links to aggregated sources, and issues corrections (three so far). He sometimes speaks like a reporter—"This weekend, I got two scoops up I was really excited about"—but also in Silicon Valley terms. He described retracting stories about startups not because they were false but as a personal favor to founders: "Founder to founder, it's like, I get it." The model depends entirely on AI agents to determine what is true, newsworthy, and legally safe—a trust that may not always hold.