Bill Swearingen develops AI patterns to evade surveillance detection

Security researcher Bill Swearingen has developed ‘noRecognition’, a project that uses adversarial machine learning to create visual patterns capable of deceiving AI-driven surveillance software.

Through approximately 31 million simulations, Swearingen generated patterns designed to disrupt the object-detection layer of computer vision systems. While the patterns appear as abstract designs to human observers, they provide enough geometric and contrast-based noise to prevent AI from classifying objects such as people, faces, or vehicles. The patterns do not blind the cameras; rather, they prevent the software from recognizing and logging the subjects.

At the DEF CON conference in Las Vegas, the technology was demonstrated using a 2009 Toyota Yaris wrapped in the specialized pattern. In collaboration with the YouTube channel Donut Media, the vehicle was driven past a Flock Safety automated license plate reader. The test successfully demonstrated that while the camera recorded the footage, the software failed to identify the object as a vehicle. Swearingen’s research also tested against other systems, including Axon body cameras and Clearview AI facial recognition software.