We Are All Playing Ben Drowned Now
When a haunted Nintendo cartridge appeared on an obscure forum thread in 2010, no one knew that sixteen years later, it would teach creators how to work with AI.
In September 2010, a college student posted a story on 4chan’s paranormal board. He’d picked up a used copy of Zelda: Majora’s Mask at a garage sale, he said. The cartridge held one save file, named BEN. He deleted it, “just the previous owner’s save file”, he thought. But when he launched the game, things got weird: textures flickered, music ran backwards, a corrupted copy of the hero followed him through warped landscapes. He uploaded the footage to YouTube and soon, it gained formidable traction. The videos were unsettling, but they felt real. Moreover, it was a time before emulating old games on a modern machine was readily available. The uploader himself swore the story was true. Within a week, thousands of people were arguing about a dead boy named Ben who lived inside a Nintendo 64 cartridge.
But every bit of it was fake.
The creator, Alexander Hall, had produced every glitch himself with a game-altering program. What he had built would later be called an ARG, an alternate reality game. But it broke the genre’s rules in one crucial way, and that broken rule is why the story matters now, to anyone who creates content with AI.
Classic ARGs present themselves as fictional. There is a puppet master, a time and place, and a fictional handshake with the participants. The players know a game has begun, even if they pretend otherwise. Hall gave his audience none of that. He presented himself as a nobody — one confused gamer among millions of uploaders — and let the footage do the talking. Researchers call this the TINAG principle: This Is Not A Game. The work hides its own nature. And when a work hides its nature, the audience’s first task changes. Before you can play, you have to discover that a game exists at all.
Get Anthony Bekirov’s stories in your inbox
Join Medium for free to get updates from this writer.
Watch what happened in the comments, because the pattern will feel familiar. One camp opted for rationalization: cartridges get corrupted, bugs happen, this kid is just superstitious. A second camp flatly called it a hoax: he’s a prankster farming attention. And a third camp believed it was real and that somehow, the cartridge was indeed haunted. Most importantly, there was no way to be sure. And this uncertainty was the experience. People replayed frames, isolated audio, hunted for hidden symbols, sometimes bordering on obsession: people ran the game’s reversed music through spectrograms looking for clues.
That was a niche thrill in 2010, a game for insomniacs on a paranormal forum. In 2026, it is simply the condition of digital media (and thus, all media). Every voice memo from a politician, every podcast clip, every unprompted DM from a stranger now carries the Ben Drowned question inside it: am I inside a game? We inspect hands in photographs, abnormal eye movements in politicians, oblong shapes in grainy pictures — every minute detail that feels amiss. There’s a word for finding patterns and intentions in ambient noise — apophenia — and it used to describe a symptom. Today it describes media literacy. The audience that Hall assembled by accident, suspicious and forensic and thrilled, has become the general audience. All of us, scanning everything, for the seams.
This is where a decade-old creepypasta has something real to offer people building with generative tools.
Here is what our research found when we studied these games closely: the doubt empowered the audience. A viewer of ordinary fiction receives a finished world. A player of Ben Drowned assembled the world personally — choosing which glitches counted as clues, which uploads belonged to the story, where the story ended and where the internet picked it back up. Each player became the editor of a private cut. The work only existed when someone performed it. That is an enormous transfer of creative agency from author to audience, and audiences loved it. They stayed for years — even after Hall revealed he created the whole thing. And they are still annotating those videos to this day.
Synthetic media has already created the doubt. The creators who thrive will be the ones who understand the second half of the equation and hand the audience something to do with it. Some of the most interesting work in AI-generated audio and video already leans this way: pieces that flag their own artificiality and invite the listener to probe it, characters that acknowledge the uncanny space they speak from, stories built to be investigated rather than consumed. The listener who interrogates what she hears is the opposite of a hater. She is the most engaged audience member you will ever have.
Which returns us to the haunted cartridge, and to the detail that separates Hall from a thousand forgotten imitators. His glitches were not random, the wrongness had craft in it, and audiences could feel this craft, even if they did not see they were being duped. Sixteen years later, the tools have changed beyond recognition and the lesson has survived intact: an audience that doubts you is an audience that’s paying attention. What you build with that attention is the game.
Adapted from academic research co-authored with Thibaut Vaillancourt (AN-ICON, University of Milan, 2023).
