Crypto
Home›Crypto›Market Structure›AI-made camouflage pattern defeats open-source surveil…
AI-made camouflage pattern defeats open-source surveillance detectors
A Kansas City researcher says the pattern fooled 11 detection algorithms used in systems such as Flock license plate readers, while video footage still records normally for human viewers.
A security researcher in Kansas City has demonstrated an AI-generated camouflage pattern that can prevent surveillance camera software from correctly classifying what the camera is capturing, including people, faces, or cars.
According to Decrypt, Bill Swearingen’s “noRecognition” project generated patterns that defeated all 11 open-source detection algorithms he tested, including software used in Flock license plate readers, Axon body cameras, and Clearview AI. In a public demo at Def Con in Las Vegas, a 2009 Toyota Yaris wrapped in one of the newest patterns drove past a Flock camera.
Swearingen told TechCrunch that the system was effective, though he said the wheels were a challenge. He reported running the same experiment repeatedly for about a year, completing roughly 31 million tests, and said he can produce patterns on demand that hide what the patterns cover from detection software.
The approach does not fully blind the camera, Decrypt reported: footage still records as normal and a human watching the screen can see a car. What is disrupted is the object-detection layer on top of the video, and the researcher described the work as adversarial machine learning that produces new patterns using a reinforcement learning model.
Decrypt added that the demo video is expected to be posted by Donut Media in the coming weeks, and Swearingen said the goal is to give people a way to opt out of being tracked by surveillance systems.