This ‘adversarial’ pattern can prevent surveillance cameras from detecting you
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Bill Swearingen ran 31 million tests over a year to develop computer-generated patterns through his noRecognition project that scramble surveillance cameras' object and face detection without blocking recording.
- The reinforcement learning model trains itself on which patterns succeed, generating new batches every minute that are "mathematically better than the last" and defeated all 11 open-source detection algorithms Swearingen tested.
- Swearingen's patterns were proven effective against software powering Flock license plate readers, Axon body-worn cameras, and cameras running Clearview AI facial recognition.
- Swearingen demonstrated the technology publicly at the Def Con cybersecurity conference in Las Vegas, covering a 2009 Toyota Yaris in one of his patterns and evading detection by a Flock camera, though the vehicle's wheels remained a challenge.
- Swearingen is withholding his strongest patterns from public release to prevent camera manufacturers from adapting their detection algorithms.
- A crowdsourcing campaign is funding early merchandise — T-shirts, hoodies, and eventually vehicle skins — to put the patterns in the hands of people who want them.
Why it matters: The tool gives citizens a way to opt out of algorithmic surveillance that powers license plate readers, AI body cameras, and facial recognition used by law enforcement, but the real-world test showed the car's wheels were still detectable — proving evasion works against software but isn't yet seamless on vehicles.
Ask SkimNews
