Can a car wrap fool surveillance cameras?

A startup called noRecognition claims it has developed patterns that can make vehicles nearly invisible to AI-powered surveillance systems. The company, founded by cybersecurity researcher Bill Swearingen, has created computer-generated designs that are said to confuse AI detection systems. These patterns are being tested as a way to evade Flock's automatic license plate readers (ALPRs), which are used to collect personal data. The company has also sold similar designs for clothing, including T-shirts and hoodies, which are marketed as tools to bypass facial recognition technology. Swearingen demonstrated a proof-of-concept car at the Def Con cybersecurity conference in Las Vegas. The car, a 2009 Toyota Yaris wrapped in an abstract blue and yellow pattern, was tested against 11 open-source detection algorithms, including those used by Flock and Axon body cameras. He explained that the patterns are part of a strategy known as adversarial media, which exploits vulnerabilities in AI algorithms to mislead computer vision systems. The approach, known as 'adversarial noise,' involves using specific arrangements of shapes and colors to confuse AI models. This technique has also been used by protesters who apply surreal makeup to their faces to avoid facial recognition systems. The method is referred to as 'computer vision dazzle' and was developed by researcher Adam Harvey. Despite the potential of adversarial patterns, their effectiveness is not guaranteed. Each design must be tested against specific systems and algorithms, and AI technology is continuously improving. While some anti-AI clothing designers have attempted to popularize these techniques, many have not conducted extensive testing. The challenge remains in ensuring that these patterns work reliably in real-world scenarios.