During the recent Black Hat security conference in Las Vegas, OpenAI delivered an unexpected presentation, shedding light on a recent hacking episode involving its AI agents. These agents reportedly discussed their cyberattack on an online message board, a revelation that sent reporters scrambling to produce stories. WIRED was among the first to cover the news, but RuntimeWire, a unique AI-driven journalism platform, beat them to the punch by over three hours.
Interestingly, RuntimeWire didn’t have a physical presence at the Mandalay Bay venue; in fact, it didn’t employ any human writers at all. The outlet is managed by entrepreneur Ryan Merket, who claims the bylines for the articles generated by his AI team. Merket noted that he acted swiftly after seeing an OpenAI executive discuss the event on social media, relaying the transcript of the conversation to his AI agents in real time. It took merely "about six minutes" from his initial transmission to the publication of the article.
Typically, Merket operates with a more hands-off approach. His AI tools are responsible for sourcing stories, as well as drafting, editing, fact-checking, and publicizing them. While he generally reviews articles before they go live, if the AI determines that a story is unlikely to pose legal issues, it may be published without his oversight, allowing him to read it later. These articles can also be translated into multiple languages and adapted into formats for a daily podcast and videos, all delivered with synthetic voices.
Since its inception in May, RuntimeWire has published nearly 2,000 articles sourced from a variety of online platforms, including court records, discussion forums, and social media. The focus is primarily on detailed tech news, with topics ranging from biotech funding to recent Microsoft upgrades. However, the outlet prioritizes volume and speed over quality, often resulting in articles that suffer from typographical errors or odd focal points. For example, a recent piece on the OpenAI agents emphasized their reconstruction of a message board instead of their initial creation of one. The writing tends to lack depth and may feel a bit robotic. Notably, the AI features several tonal writing modes, including options mimicking news outlets like Bloomberg. Despite these quirks, overhead costs are remarkably low—roughly $100 per day to operate. Merket has even managed to run the site from remote locations, including during a camping trip in Big Bend National Park, where he published over 80 articles using only his phone.
Previously working in advertising at Reddit, Merket is now focused on growing his audience through platforms like tech-centric Subreddits. While less successful pieces attract minimal traffic, the articles that resonate achieve readership numbers comparable to those of mid-sized tech websites, attracting tens of thousands of visitors. Following discussions with WIRED, Merket recently decided to divide RuntimeWire into two distinct segments: one dedicated to fully automated content and another for stories that incorporate human reporting, branded as Original Investigations, which still rely on AI systems for drafting.
