Adversarial Fashion Confronts Surveillance Norms

Garments engineered to interfere with AI-powered surveillance cameras are moving from art projects into a small commercial industry, according to a report on countersurveillance fashion. The trend was highlighted by noRecognition, a Kickstarter project presented last month at the DEF CON hacker convention.

In 2025, cybersecurity expert Bill Swearingen began experimenting with a Python-based fuzzer targeting YOLO, a popular object detection framework. He then developed a reinforcement learning algorithm that generates colorful geometric adversarial patterns, which he tested against 11 object detection models: four that search faces, two that recognize faces, and five that detect people, most publicly available. Successful patterns lowered the models' confidence scores, sometimes to the point of no detection. "Privacy is a human right," Swearingen said.

Two companies already sell physical garments. Cap_able uses a patented method to weave bright motifs into jacquard knitted fabrics; founder Rachele Didero, an assistant professor at the Free University of Bozen-Bolzano, said the clothing interferes with certain computer vision systems, particularly those backed by fast convolutional neural networks, potentially leading them to classify wearers as animals or objects. Urban Privacy's Faception Reloaded collection uses black-and-white prints abstracted from a human face that appear as additional faces on detectors, slowing them down; co-founder Daniel Preuß said the idea is "to create false data." Its asymmetrical cuts and wide silhouettes intend to conceal body shape and gait.

Experts cautioned about effectiveness. Niloofar Mireshghallah, incoming professor at Carnegie Mellon University, said real-world conditions such as camera angles, lighting and fabric folds can reduce performance, and that one good frame is all a system needs. Patterns must also be tuned to specific models, and operators could train future models on a given pattern and wearer. Dippu Kumar Singh of Fujitsu North America called the clothing a fragile shield against a constantly improving threat. Preuß said it is "not an invisibility cloak."

Cap_able plans further innovation, and Urban Privacy will release more items including a "shadow cap" with an acrylic face shield and cutouts. Swearingen plans to explore anomalies he encountered, including a pattern that shifted a bounding box and another that changed a camera setting. Mireshghallah described countersurveillance fashion as a speed bump and warned that aggregating weak signals is the real risk.