x
Automated surveillance has quietly reshaped the modern landscape. From automated licence plate readers mounted on light poles to body-worn cameras and facial recognition software scanning public squares, artificial intelligence is constantly logging, categorising, and tracking human movement.
For privacy advocates, the unchecked expansion of these networks presents a daunting challenge. However, a novel countermeasure has emerged from the cybersecurity research world: algorithmic camouflage.
By exploiting the mathematical vulnerabilities of computer vision models, security researcher Bill Swearingen has created a way to turn human clothing and vehicle wraps into active cloaking devices against AI detectors.
Developed by Kansas City security researcher Bill Swearingen (co-founder of the SecKC security meetup), noRecognition is an open-source research initiative designed to generate adversarial patterns on demand.
Unlike traditional camouflage, which aims to blend an object into its natural physical background, Swearingen’s patterns aim to confuse the digital layer operating on top of video footage.
When a surveillance camera records a scene, two distinct processes take place:
The patterns produced by noRecognition completely disrupt this secondary classifier layer. The camera continues to capture clear video footage, but the automated system fails to recognise that an object or individual is present. In essence, it prevents your data from being automatically indexed into a surveillance database.
To understand how a brightly coloured, geometric pattern can trick a multi-million-pound surveillance network, one must look at how computer vision operates.
Computer vision models do not interpret images the way human eyes do. Instead, they process pixel values, gradients, edges, and mathematical feature maps. By injecting specifically engineered "visual noise" into a pattern, developers can trigger mathematical contradictions within the algorithm's neural network.
Swearingen built a reinforcement learning model that essentially grades its own output:
After running approximately 31 million tests, Swearingen’s system learned how to "paint" against machine logic. The resulting patterns defeated all 11 major open-source detection algorithms tested, including systems that underpin widely deployed technologies such as Flock licence plate readers, Axon body cameras, and Clearview AI facial recognition software.
The theoretical capability of adversarial noise was brought into the physical world at the Def Con hacker convention in Las Vegas. Collaborating with the popular automotive YouTube channel Donut Media, Swearingen wrapped a 2009 Toyota Yaris in one of his newest AI-evading patterns.
The vehicle was driven directly past a live Flock surveillance camera to measure its real-world performance. Despite the challenges posed by uncovered tyres and vehicle glass, the test successfully proved that engineered visual noise could stop automated systems from logging the car as a vehicle entry.
To prevent camera vendors and law enforcement suppliers from using his output to retrain their algorithms, Swearingen keeps his strongest, most effective patterns offline. Because the model continuously generates fresh designs, the evasion software evolves faster than static detection models can adapt.
The motivation behind noRecognition stems from a growing public desire to maintain personal anonymity in an era of ubiquitous digital logging.
Over recent years, automated licence plate readers have drawn criticism from privacy organisations and lawmakers alike over concerns regarding mass data collection, wrongful stops due to bad algorithm matches, and potential overreach by federal agencies.
While citizens have historically relied on basic physical methods to guard their identity—such as wearing masks, brimmed hats, or hoodies—adversarial clothing and vehicle skins represent a sophisticated technological shift. They allow individuals to go about their daily lives in public spaces without being automatically catalogued into centralized AI dragnet databases.
While adversarial apparel like T-shirts and hoodies can be worn freely, applying these patterns to transport brings complex legal questions.
Licence plate obstruction laws vary widely across jurisdictions. Swearingen’s vehicle wraps are deliberately engineered to cover the vehicle's bodywork rather than the licence plate itself, testing the legal boundaries between standard vehicle styling and intentional automated evasion.
Through a crowdfunding campaign aimed at funding high-resolution apparel and vehicle skins, the noRecognition project aims to make privacy-preserving designs accessible to the wider public. As surveillance algorithms become more pervasive, engineered visual patterns may soon transition from a niche cybersecurity experiment into a standard tool for digital self-defence.
Disclaimer: This article is provided for informational purposes only, mistakes may be made, and it's not offered or intended to be used as legal, tax, investment, financial, or any other advice.
