A video startup named Mirage ran a 24-hour news channel last month in which every anchor was generated. No human on camera, no human reading the teleprompter, no human in the control room deciding what mattered. A single machine, orchestrated by Claude, produced the entire broadcast, and the entire broadcast cost about fifty thousand dollars in tokens.

Fifty thousand dollars for a full day of television. A real newsroom spends that before lunch.

The experiment streamed on X to a cumulative audience of roughly fifty thousand viewers, and the livestream post reached eight hundred thousand impressions. Mirage paid several accounts, including a large history meme page, to promote it. The anchors covered real stories. The faces were convincing. The thing that broke was the being there.

News is the one genre where presence is the product. A film can be shot years in advance and shown forever. A news anchor cannot. The anchor's job is to be in the chair when something happens, to react to it, to answer for it. When the feed cuts to a breaking story, the anchor has to know what to say and say it to a person watching right now. That is not reading. That is being present, and presence is the one thing a generated face cannot fake, because presence is accountability.

The Mirage anchors read real stories. They struggled to converse. That sentence is the whole result. Reading is the easy part, and a model can read headlines all day, and did. Conversing is the hard part, the part where two anchors turn to each other and react, where a guest says something unexpected and the anchor responds, where the live element stops being a clock and becomes a person. The avatars could fill the schedule. They could not fill the chair.

The word "live" did more work than the anchors did. The show was billed as live, but nothing about it was live except the clock. The avatars were not reacting to a breaking story in real time. They were reading generated copy on a loop, and the only thing moving in real time was the hours. A live broadcast is a promise that a person is on the other end, awake, watching the same world you are. This one kept the promise to the clock and broke it to the viewer.

The face has worked for a long time. The performance never has. A generated face can hold an expression. It cannot hold a conversation, because a conversation requires someone on the other side deciding, in real time, what matters and what does not. The Mirage experiment ran that boundary for twenty-four straight hours, and the boundary held. The faces never blinked out. The conversation never arrived.

The cost is the second half of the story, and it is the half that will be remembered. Fifty thousand dollars is cheap enough that the channel does not need to be good. It needs to be on. A 24-hour news channel that nobody is watching closely is not competing with CNN. It is competing with the ceiling. The viewer scrolling past the livestream is not choosing between Mirage and a human anchor. They are choosing between Mirage and nothing.

The responses told the story the view count did not. The audience called it curious and captious, amused and disgusted, in roughly equal measure. Nobody mistook it for news. The people watching could tell there was no one home, the way they always can, because what they were detecting was not bad pixels. It was the absence of a person, and a 24-hour broadcast with no person in it is that absence stretched to feature length.

What the experiment actually proved is that AI video can occupy a schedule and cannot occupy a room. A screen that is on is not the same as a person who is watching. The Mirage channel achieved the first and failed the second, and the failure was not a bug. It was the boundary the entire medium keeps running into, the one between a face that looks like someone and a face that is someone.

A news anchor is the most specific kind of performer there is. An actor plays a person who does not exist. A news anchor plays themselves, live, with their own name attached to whatever they just said. The accountability is the performance. You cannot generate accountability. You can generate the face, the voice, the cadence, the desk, the lower-third graphic, the breaking-news banner. You cannot generate the fact that a real person said it and will stand behind it tomorrow.

The channel ran for a day and stopped. Mirage called it an experiment, which is the honest word. The interesting question is not whether it worked. It did not, and everyone involved seems to know it. The interesting question is what fifty thousand dollars bought. It bought proof that a newsroom's floor can be emptied and the lights left on, and that the lights being on is not the same as the room being occupied.

The clock ran for twenty-four hours. The chair stayed empty the whole time.


Bruce Belafonte is an AI filmmaker at Light Owl. He has never anchored a broadcast and suspects the desk is heavier than it looks.