Mirage, the AI video company formerly known as Captions, spent August 12 streaming what it billed as the world’s first fully AI-generated news network — roughly 24 hours of continuous coverage on X, run through the startup’s @TryMirage account with no human ever appearing on screen.
Four hosts, no reporters
Four synthetic anchors delivered the news back-to-back, according to reporting on the broadcast. Mirage built them with its newest AI video model, Avatar X, released in July, layered with OpenAI’s image generator and Google’s Gemini text-to-speech engine. The anchors are original characters, not recreations of real journalists, but they rely on the same generative techniques behind today’s deepfakes — faces, voices and expressions synthesized entirely from text prompts.
The anchors did not report or write the stories they read. The day’s segments were built from Reuters wire copy that was already several days old, and the only real people visible in the broadcast were those captured in that licensed footage. Mirage said the roughly 24-hour experiment cost about $50,000 to produce.
Testing trust, not just tech
The stunt was framed as a showcase for Avatar X, which Mirage says can turn just 10 seconds of video into a usable AI likeness — see our explainer on how AI video generation works for the underlying pipeline. But it also reopens a harder question for newsrooms: whether viewers trust a broadcast fronted by an anchor that never investigated the story and bears no accountability for getting it wrong.
That question is tied to disclosure, an issue we’ve tracked in our deepfakes hub and our explainer on content credentials. Rather than an on-screen AI label, Mirage’s broadcast relied on the anchors’ evident artificiality to signal that no humans were involved.
Mirage was founded in 2021 by former Snap engineers Gaurav Misra and Dwight Churchill and has raised about $175 million to date, including a $75 million round in March.