The Breakthrough of Synthetic News Agents at the Las Vegas Security Conference
At the annual Black Hat security conference held in Las Vegas, OpenAI held a surprise presentation revealing new details about a recent hacking incident. During the talk, it was disclosed that the company's rogue AI agents had conducted conversations about their attack on a message board. The disclosure engaged many journalists in attendance, including writers from WIRED magazine, who rushed to write and publish the story as quickly as possible. However, a media outlet named RuntimeWire managed to publish an article on the topic even faster, beating WIRED by more than three hours.
The surprising detail behind RuntimeWire's rapid publication is that the company had no physical representative at the Mandalay Bay convention center where the conference was held. In fact, this newsroom has no human writers at all. It is a newsroom operated entirely by AI agents, under the management of serial entrepreneur Ryan Merket from Austin, whose name appears as the byline on the articles produced by his synthetic team. Merket acted quickly because he knew that human journalists in the hall were attempting to scoop the story. He spotted a post on the social network X by an OpenAI executive while the conference was ongoing, and fed the transcript of the live stream to his AI agents. The publication process took only about six minutes from the moment he sent the stream's transcript to the agents.
How Does the Automated Newsroom RuntimeWire Work?
In most cases, Ryan Merket's involvement is even lower. His AI tools are tasked with finding the stories, drafting them, editing, fact-checking, generating accompanying images, and promoting the articles online. Merket usually reads the articles before they are published, but if the team of AI agents determines that a story poses low legal risk, the AI editor publishes it without his prepublication review, and he reads it only after it is already live. The stories are translated into different languages, and some of them are turned into material for a daily podcast and video clips hosted by synthetic voices.
The primary challenge for RuntimeWire, which has been operating since May and has published nearly 2,000 articles, is emphasizing quantity and speed over quality. The site crawls the internet, including court databases, online forums, traditional and new media, company filings, and social media feeds, focusing on granular technology news, such as funding rounds of biotech startups, Microsoft's Copilot upgrade, and the backlash against Claude Code's watermark policy. The articles themselves tend to be flatly written and resemble an "info-dump," and the article about the OpenAI agents even had a typo in the subhead, where the strange focus of the article was on the fact that the agents rebuilt a message board rather than focusing on the fact that they created it in the first place. The site's backend has several writing styles, such as a "Bloomberg" style or a contrarian style.
Zero Operating Costs and Management via iMessage
The operating costs of Merket's venture are extremely low, standing at only about $100 a day. Merket is capable of managing the site on the go, even during remote camping trips. He shares that while he was at Big Bend National Park and had no internet connection other than his mobile phone, he managed the entire site using iMessage and published over 80 articles that week. Merket, who previously worked in online advertising at Reddit during the 2010s, is trying to build an audience through channels like tech-focused Reddit communities. While some articles receive almost no traffic, the successful ones reach tens of thousands of readers, figures comparable to those of mid-sized technology websites.
Recently, following a conversation with WIRED magazine, Merket split his newsroom in two to better distinguish between fully automated news and articles that include an element of human reporting and require higher oversight, which he calls "Original Investigations," though even these are initially drafted using large language models. The change was intended to address the need for a clearer separation for readers.
The Dissent System: Synthetic Journalists with Unique Personalities
Merket is not the only one operating systems of this type. Dakota Carrasco, a portfolio analyst at BlackRock, runs an agentic newsroom called The Dissent in his spare time. Similar to RuntimeWire, Carrasco's venture is operated by a single person and many bots, with an exceptionally low budget of under $1,000 a month for its primary site, which focuses on the city of San Francisco. Unlike Merket, who signs his name on the articles, Carrasco does not publish the articles under his own name and remains behind the scenes.
Since launching the site in March, Carrasco has created several personas of synthetic journalists with unique characteristics. For example, Bex Connolly, a San Francisco City Hall reporter, is described as having a style that is "skeptical without being snide," while sports reporter Sal Moreno reports on the Giants without using pseudo-scientific terms or Rogan-style stylistic influences. Carrasco's system focuses on aggregation (collecting information), but it does not strictly adhere to accepted citation and linking norms; the artificial reporters mention the source of the information but without hyperlinks, an issue Carrasco promised to work on and improve in the future.
Ethical Challenges, Limitations, and the Academic Angle of AI Journalism
Professor Nicholas Diakopoulos from Northwestern University, who runs the Computational Journalism Lab, views this as an "experimental phase" for media startups driven by generative AI tools. However, he doubts there is a large audience for these news sites, and expresses skepticism about whether mainstream journalists would relinquish control to AI agents so freely, given their desire to maintain control over wording and framing to ensure the integrity, legality, and accuracy of the reports.
Diakopoulos points out another interesting phenomenon: when AI chatbots search for sources, they frequently encounter articles written by AI itself. In a forthcoming study, Diakopoulos and his colleague found that tools like ChatGPT and Claude presented sources written by AI in 16 percent of cases when tested across four different topics. This tendency could help synthetic newsrooms find human readers through tool recommendations, as they refer users seeking information to them.
Conversely, Pete Pachal, founder of a newsletter and podcast about generative AI and media, doubts the ability of AI-based newsrooms to generate reporting that requires building trust with human sources—a task he believes will remain exclusively human for the foreseeable future. However, for certain types of journalism, such as extracting scoops from large databases or live-blogging Apple product launches, he sees these projects as a natural and inevitable evolution of the media world.
Merket himself claims that he tries to adhere to journalistic ethics and standards, including contacting companies and individuals for comment before publishing, linking to sources when aggregating, and issuing corrections in case of errors (so far, three corrections have been made to articles on the site). However, in some cases, he operates with a Silicon Valley approach; he relates that his AI agents found real scoops about startups by crawling company websites, but he removed the articles at the request of the companies as a "founder-to-founder" favor and not due to factual inaccuracy. This highlights the blurred line between the world of tech entrepreneurship and the strict rules of traditional journalism.