Is there an audience for AI-delivered news? Public data says no, and previous attempts to deliver such ventures have yielded many human-written stories about projects that never materialized.

And yet Mirage — the AI startup behind the video-editing app Captions that raised $75 million this year and was last valued at $500 million — thinks someone, somewhere out there may have a hunger for an AI avatar to supplant their cable news habits but just hasn’t had an outlet to indulge in it.

Enter the “Mirage News Network,” the company’s daylong experiment this week to create a 24-hour newscast where AI-generated news anchors cover real stories. The broadcast reached a total of 50,000 viewers on X, with its livestream post reaching 824,000 impressions and drawing responses ranging from curious to captious — while Mirage paid several accounts, including @historyinmemes, to promote the livestream.

The four anchors were completely generated by an AI agent using a mixture of Mirage’s Avatar X model, which the Mirage News Network release was ostensibly launched to promote, and tools like ChatGPT’s image generator and Google’s Gemini Text-to-Speech feature. The days-old stories were licensed from Reuters’ wire service, and the only real people to appear across the stream were those featured in the videos that Reuters journalists filmed and the stock footage that accompanied such packages. For AI-generated guests who appeared for “interviews,” Mirage recruited more than a dozen people, mostly venture capital professionals, who authorized their likenesses and approved the scripts for what their likenesses would say. (One notable name: Billy McFarland, of Fyre Festival and federal prison fame.)

Mirage CEO Gaurav Misra, who built and managed the project by himself with $30,000 worth of Claude Code, wanted the newscast to feel like a traditional broadcast, drawing inspiration from ’80s-style newscasters’ looks and appearances with names like “Tom Callahan” and “Marcus Sterling.” His mission, Misra told Variety, was to figure out whether there was an appetite for something that resembled the news but where nothing but the stories themselves was rooted in reality.

“The goal of this was not exactly to be the news or to fool anybody, or to pretend to look real,” he said. “In fact, the goal was to make sure that people know that this is AI-generated. That was the most important part.”

Yet Jason Silberman, a software engineer at Mirage, may have given away the game in an X post: “when the CEO starts tokenmaxxing too hard and spends $50k in a day on a marketing stunt.” “Tokenmaxxing,” according to one definition, refers to “maximizing AI token usage — and treating that volume as proof of productivity.”

Still, Misra’s points touch on a foundational element of ethical journalism, which the Society of Professional Journalists’ code of ethics demands be “accountable and transparent” in the work. But such codes did not power the Mirage project — nor did any journalists who would adhere to such rules, because no journalists work at the company and it is not a news organization. 

Misra’s background is not in news but in tech, having worked as a software engineer at Snap and Microsoft. His instructions to the machine were not to follow any specific ethical code, guidelines or style, such as the journalistic bible of AP style or the widely accepted SPJ Code of Ethics, but instead to adhere to “general journalistic standards,” Misra said. The product represents another example in the AI era of a group of people — in this case, fewer than 10 — working on a project within a specific industry without a perspective informed by years of experience in that field.

On one hand, the operation was transparent. The livestream featured a graphic in the top-left corner signifying nearly everything on the screen — the anchors reading the story scripts, the graphics accompanying such stories, the “guests” who appeared for interviews and even the commercials between segments  — was artificially created. When the program made errors — and it did, repeatedly, such as incorrectly identifying a woman in a Reuters package as President Donald Trump — the company disclosed such errors on its X account. (Though it did not, as most TV news outlets do, do that during the broadcast itself; Misra watched the livestream hours ahead of the broadcast to try and catch errors before they aired.)

But such a lack of experience in news or specificity in journalistic standards in working on a project like this on an experiment such as this represents “a false and misguided devaluing and deskilling of journalism,” said Dan Axelrod, a former journalist who currently chairs the SPJ’s ethics committee, who called the endeavor not a news network but “a news delivery project” that was not “creating a product that adds value to the marketplace.” The SPJ ethics committee this week introduced a draft revision of its code for the AI age, designed to press journalists to adhere to their ethical obligations as technology adapts. 

“Nobody says, ‘Let me go do my own surgery,’” Axelrod said. “Few people say, and many immediately come to regret it, ‘Let me defend myself in court.’ Everybody thinks they can communicate, or be the best communicator, or be the best storyteller, and they don’t take into account all of the training and all of the experiential learning and all of the ethical guidance that a high-level professional journalist follows, and they don’t take into account or value [it].”

Axelrod said the project betrayed journalistic credibility because such work “hinges on the simple promise that a human with professional judgment stands behind the story, and the news unites in that it captures and connects us by our humanity.” 

“We watch it to see and feel the things that other humans see and feel,” Axelrod added, “and that goes for the journalists, not just the people covered in the stories.”

News was the first industry Mirage tackled because he didn’t believe it could be replicated through AI. But now that the company has produced this, Misra thinks there’s a world where AI-generated newscasts are “probably inevitable in the future” bcause AI “seems to be surprisingly good at just stating the facts, stating the facts in a way that doesn’t even try to hint towards which one is the right answer, which is which is not easy, or at least it seems to be not easy.”

Misra’s belief in how journalism should be delivered hinges somewhere along recognizing “a battle between being completely dry and accurate versus actually something that people want to watch” while believing people opting for opinion-based content that was “more entertaining, probably.” His news diet consists of small bits of TV news, Google News aggregations and for “the rest of it, I depend on TikTok.”

But Misra believes there may be an audience for such endeavors, “everyday people” who he thinks may prefer a generated person to a living, breathing anchor for “a different angle” for news. It also, he contended, “just lowers the barrier for more people to be able to participate in in creating content and things like that.”

“But I don’t think it’s necessarily true that you know people want one thing or the other,” he said. “I think different people might be happy with different things.”

According to a Gallup survey conducted in May 2026, only 7% of Americans either rely “a great deal” (2%) or “a fair amount” (5%) on AI to get news, but nearly 40% of Americans polled believe that a news organization’s use of AI would “reduce my trust in the information.” Similar figures were produced by the Pew Research Center last year regarding AI chatbots, with nearly 60% of adults between ages 18 to 29 and 51% of adults between 30 and 49 believing at least some of the news produced by chatbots is inaccurate. In June, Pew found that 40% of respondents believe AI would ultimately be worse for society.

It raises a question that only “Stephanie Kwan,” the “tech desk” “reporter” for the “Mirage News Network” could pose, as the AI avatar did for a Reuters package about robot dogs replacing security guards in Atlanta.