Adversarial pattern blocks surveillance cameras from detecting people

Bill Swearingen, a cybersecurity professional based in Kansas City, has spent the past year running roughly 31 million tests to develop computer-generated patterns that block surveillance cameras from detecting whatever they cover. His project, called noRecognition, uses a reinforcement-learning model that trains itself on which visual patterns defeat specific camera detection algorithms — a process Swearingen describes as teaching the model how to paint.

Swearingen, who co-founded the Kansas City cybersecurity meetup SecKC, said the patterns don’t stop cameras from recording; instead, they scramble a camera’s ability to identify the people, vehicles or objects they cover, so no detection alert is triggered. The model can now produce patterns that defeat all 11 open-source detection algorithms Swearingen tested, including the software that powers Flock license plate readers, Axon body-worn cameras, and Clearview AI. It now generates new patterns every minute, with each batch improving on the last, he said.

Swearingen said he built the tool because his hometown is covered in surveillance cameras he never consented to, comparing it to how the government uses his driver’s license photo for facial recognition. He recalled feeling uncomfortable attending a protest last year because of how many cameras could track participants exercising their right to free expression. “Privacy is a fundamental right,” Swearingen told TechCrunch, describing the patterns as a way for people to stop being tracked.

On Friday, at the Def Con cybersecurity conference in Las Vegas, Swearingen ran the project’s first public, real-world test: with help from Donut Media, he covered a 2009 Toyota Yaris in one of his newest patterns and demonstrated that it could defeat detection by a Flock camera, though he said the car’s wheels remained a challenge. Donut Media said video of the demo will be released in the coming weeks.

noRecognition is running a crowdsourcing campaign to fund early merchandise, including T-shirts and hoodies printed with the patterns, and Swearingen said there is potential for pattern-printed vehicle skins down the line. He said he is withholding his strongest patterns from public release so camera makers can’t train around them, noting that every failure only makes the underlying model improve further.

Why it matters

The project is early evidence that individuals can counter algorithmic surveillance without any change in law or policy, at a moment when license-plate readers and facial-recognition cameras are proliferating across US cities. If the approach scales beyond a hobbyist project, it could complicate law-enforcement and commercial surveillance systems that rely on automated detection rather than human review.