To the human eye, the Digital Camouflage shirt has an abstract colorful pattern showing green, pink, and orange shapes. To robots, it hides the human wearing it.
German designer Simon Weckert created the button-up shirt to confuse AI surveillance systems following Berlin’s installation of surveillance cameras earlier this year. The garment is part of an emerging trend of adversarial clothing design that uses complex patterns to disrupt facial and object recognition.
“Object detection systems don’t ‘see’ the way we do,” Weckert tells Fast Company. “They scan an image for statistical patterns—combinations of edges, textures, and colors that they’ve learned to associate with the category ‘person.’”
The pattern was mathematically optimized to send the wrong signals to these networks. Available for about $88, Weckert says his shirt has to pull double duty: confound AI but also still be wearable.
“The raw adversarial patterns that come out of the optimization are effective but visually chaotic—pure machine-facing imagery,” he says. The final pattern sits in the middle ground between algorithmic needs and human taste, and he says he finds the ambiguity conceptually important.
“It’s a garment addressed to two kinds of viewers simultaneously, and each sees something completely different,” he adds.
Weckert tested the pattern against You Only Look Once (YOLO), an open-source detection system built on the same class of technology that powers many commercial surveillance products. He notes, however, that the garment was not necessarily designed to defeat a specific surveillance system, but rather to raise questions about how much trust a technology deserves if a simple piece of fabric can undo it.
“I make no claims about specific proprietary systems,” he says. The point of the shirt isn’t to defeat one product, but to show a structural fragility in this entire class of technology.