Browse
Log in / Register
Share
March 12, 2025 – A U.S. security researcher has developed a computational pattern that disrupts automated image recognition systems, raising the possibility of both evading mass surveillance and protecting individuals from camera-equipped smart glasses. The pattern, called “noRecognition,” was created by Bill Swearingen and unveiled at the Def Con hacker conference in Las Vegas. In a live demonstration, Swearingen applied the pattern to a 2009 Toyota Yaris and tested it against a Flock camera—a device commonly used for automated license plate and vehicle recognition. The system failed to identify the wrapped vehicle as intended, though the wheels remained a weak point.
Instead of making a person or object invisible, noRecognition exploits the weaknesses of machine learning models. The pattern—a non-repeating, geometrically chaotic design—is printed on fabric or vehicle wraps. To the human eye, it looks like abstract art. But to AI models trained to detect faces, vehicles, or license plates, the pattern introduces enough noise to prevent reliable classification. Swearingen ran approximately 31 million tests over a year, optimizing the pattern against eleven open-source recognition models, including those used for license plate reading, body cameras, and facial recognition.
The key distinction is between recording and algorithmic recognition. A surveillance camera will still capture the video; a human operator can still see the person or car. But the AI that tags, searches, or triggers alarms will likely fail to process the object. This makes noRecognition a tool for defeating automated analysis, not stealth.
At Def Con, Swearingen fitted a 2009 Toyota Yaris with a pattern designed to confuse vehicle detection. He drove it past a Flock camera, which is deployed by law enforcement and homeowners associations to identify vehicles. According to TechCrunch, the test worked in principle: the camera did not recognize the car as a known vehicle. However, the wheels remained uncovered, and Swearingen acknowledged that the pattern's effectiveness may vary with different lighting, distances, angles, and software updates.
Independent large-scale testing has not yet been conducted. The pattern has only been validated against open-source models, not the proprietary systems used by companies like Flock, Clearview AI, or major surveillance vendors. Critics note that adversarial patches have been studied for years in academia, but translating them into a reliable, mass-produced product remains a challenge.
Through a crowdfunding campaign, Swearingen plans to bring noRecognition to everyday items: T-shirts, hoodies, scarves, and eventually vehicle wraps. The potential application for smart glasses is particularly timely. As camera-equipped glasses like Meta Ray-Ban, Apple’s rumored smart glasses, and others begin to incorporate AI analysis of surroundings, a pattern that disrupts object detection could serve as a privacy safeguard.
However, the current noRecognition designs have only been tested against a limited set of recognition systems. Whether they are effective against the specific camera sensors and AI models embedded in smart glasses—often smaller, lower-resolution, and running on-device machine learning—is unknown. Swearingen has not disclosed any tests with such devices. The pattern may also be less effective against systems that use multiple sensor modalities (e.g., depth, infrared) or that are trained on adversarial examples.
NoRecognition is not the first attempt to fool AI with physical patterns. Academic research has produced adversarial patches that can hide a person from a YOLO-based detector or cause a stop sign to be misclassified as a speed limit sign. What sets Swearingen’s work apart is the focus on scale and commercial viability: he aims to produce wearable items that regular consumers can purchase.
Yet experts caution that such patterns are often brittle. “The history of adversarial examples shows that what works in a lab under controlled conditions can fail in the real world,” says Dr. Helena Torres, a computer vision researcher at MIT. “Lighting, occlusion, and the specific model version all matter. A pattern that confuses an open-source model may not work on a proprietary one that has been trained to be robust.”
The development of tools to evade AI surveillance raises complex questions. Privacy advocates argue that individuals have a right to opt out of mass automated identification, especially in public spaces where consent is absent. The Electronic Frontier Foundation (EFF) has long supported research into “counter-surveillance” technologies. Others worry that such patterns could be used by criminals to hide from legitimate law enforcement cameras.
Swearingen’s crowdfunding campaign states that noRecognition is intended for “privacy-conscious individuals, not for illegal activities.” The project’s website includes a disclaimer that the patterns should not be used to evade security checks at airports or government buildings.
The next step for noRecognition is a larger-scale field test with multiple camera types and environmental conditions. Swearingen is also exploring dynamic patterns that could change based on the camera view. For smart glasses, the pattern may need to be redesigned to work with smaller, closer lenses.
As of March 2025, noRecognition remains a promising but unproven technology. Its success will depend on whether it can withstand the adversarial arms race: as surveillance AI improves, so must the countermeasures. For now, the pattern offers a glimpse of a future where individuals can visually announce, “Do not recognize me.”
