Researcher develops AI-generated pattern to evade surveillance cameras

Cybersecurity researcher Bill Swearingen has developed the noRecognition project, which uses a self-trained reinforcement learning model to generate adversarial patterns that prevent surveillance cameras from detecting people and vehicles. Tested over 31 million times, the patterns successfully scramble the software layer of 11 major detection algorithms, including Flock and Clearview AI, without blocking the video recording itself. Swearingen and Donut Media demonstrated the technology's real-world efficacy at the Def Con conference in Las Vegas by covering a 2009 Toyota Yaris in the pattern to successfully evade a Flock camera.

noRecognition adversarial pattern project

• ▪The noRecognition project launched a crowdfunding campaign to sell early merchandise featuring the adversarial patterns, including T-shirts, hoodies, and potential vehicle skins.
• ▪Bill Swearingen developed the noRecognition project to create computer-generated patterns that prevent surveillance cameras and license plate readers from detecting people or vehicles.
• ▪Bill Swearingen is keeping his strongest adversarial patterns off the internet to prevent surveillance camera manufacturers from developing defenses against them.

Surveillance camera detection evasion

• ▪The adversarial patterns scramble a surveillance camera's ability to identify objects, people, or faces to prevent triggering detection alerts, without blocking the camera from recording video footage.
• ▪The adversarial patterns rely on adversarial machine learning, exploiting the fact that computer-vision systems interpret images differently than humans to cause detection models to misclassify or fail to identify objects.

Reinforcement learning model development

• ▪The reinforcement learning model successfully generated patterns that defeated all 11 open-source detection algorithms tested, including software powering Flock license plate readers, Axon body cameras, and Clearview AI.
• ▪Bill Swearingen developed a reinforcement learning model that trained itself by running approximately 31 million tests over a year to refine patterns against specific camera algorithms.

Def Con real-world demonstration

• ▪On August 7, 2026, Bill Swearingen conducted the first public real-world test of the noRecognition pattern at the Def Con cybersecurity conference in Las Vegas.
• ▪The Def Con demonstration, conducted with Donut Media, involved covering a 2009 Toyota Yaris in an adversarial pattern to prove it could successfully evade detection by a Flock camera.

Privacy rights advocacy

• ▪Bill Swearingen was motivated to start the project due to the high density of surveillance cameras in Kansas City and his personal discomfort with potential camera tracking while attending a protest in 2025.
• ▪Bill Swearingen, a Kansas City-based cybersecurity professional and co-founder of SecKC and SIXCYBER, stated that privacy is a fundamental right and described his patterns as a way to opt out of tracking.

Automated tracking concerns

• ▪The automated detection algorithms used in modern surveillance cameras allow law enforcement to sift through footage to identify specific activities, license plates, or faces.
• ▪Law enforcement agencies, private companies, and local governments widely deploy automated detection tools to process vast amounts of footage, though the systems have drawn scrutiny over errors and inaccurate matches.