As AI-powered cameras increase in prevalence, a new industry is developing "adversarial fashion" to disrupt surveillance. Projects like noRecognition, presented at the DEF CON hacker convention, and brands such as Cap_able and Urban Privacy, are crafting garments specifically designed to evade or baffle computer vision systems.

Cybersecurity expert Bill Swearingen developed a reinforcement learning algorithm to generate adversarial patterns. These colorful geometric designs were tested against 11 object detection models—including those for face and person detection—and successfully lowered their confidence scores, sometimes preventing detection entirely.

Cap_able produces garments using a patented jacquard knitting method. Their clothing aims to interfere with fast convolutional neural networks, which are a type of artificial neural network used for image processing, potentially causing the system to classify the wearer as an animal or an object instead of a human.

Similarly, Urban Privacy's "Faception Reloaded" collection uses black-and-white prints of abstracted human faces to create "false data" for facial recognition algorithms. While these patterns can slow down detectors, experts note significant limitations. Factors such as camera angles, lighting, garment folds, and gait recognition can reduce effectiveness. Furthermore, machine learning models may eventually be trained to recognize and bypass these specific adversarial patterns.

Sources: