Decoy Font hides two messages in the same letters: a human reads one, a language model reads the other. It's built on an optical illusion decades old. The author himself says the trick may not last long.
- The typeface has two layers: thin, clear outlines up front and blurred dark shapes behind; the human sees the dark ones, the model reads the outlines.
- In the demo, an image that reads 'sleep early' to a human is read by two leading models as 'binge shows.'
- The technique is the familiar hybrid image; the background letters are built on the free typeface DejaVu Sans Mono.
Once every few months something comes along that's craft, joke and statement at the same time. This is one of those.
I've made logos that had to work both up close on a business card and from the far end of a billboard - the shape reads differently at every distance, and any good designer knows it by heart. Decoy Font plays exactly that trick, only flipped around, against a machine instead of a passerby's eye. The weapon here isn't new technology - the hybrid image is an old trick of perception, the same one behind the famous portrait that swaps faces depending on distance. Something known for decades gets pointed at the newest thing on the market. That's good craft: you don't invent a world, you find the gap in its vision.
There's a second layer I like even more than the typeface itself. The author isn't selling protection. He says outright that in a few months the models will probably read this without trouble, and that the value is in tracking the change. Rare honesty in a field where every other project calls itself a solution.
Still, I wouldn't recommend it as a way to keep your text private - whatever passes through someone else's server gets read. What matters here is different: it shows there's still a gap between human and machine vision, and that the gap is material a designer can work with. AI fatigue is real, and people are looking for a way to answer it. Sometimes the answer is a typeface.