Briefly
- Invoice Swearingen’s noRecognition undertaking generates patterns that cease digital camera software program from classifying what it covers—individuals, faces, or automobiles.
- The patterns defeated all 11 open-source detection algorithms he examined, together with the software program behind Flock license plate readers, Axon physique cameras, and Clearview AI.
- The primary public take a look at got here Friday at Def Con in Las Vegas: a 2009 Toyota Yaris wrapped within the sample, pushed previous a Flock digital camera.
Invoice Swearingen spent the previous yr working one experiment again and again from his dwelling in Kansas Metropolis, the place he co-founded the SecKC safety meetup. About 31 million checks later, he says he can produce patterns on demand that disguise no matter they cowl from the detection software program wired into Flock cameras—the controversial surveillance system being rolled out throughout America.
He confirmed it in public for the primary time Friday at Def Con, working with the YouTube channel Donut Media to cowl a 2009 Toyota Yaris in one among his latest patterns and roll it previous a Flock digital camera.
“We proved it was efficient,” Swearingen told TechCrunch, although he stated the wheels have been a problem. Donut Media stated video of the demo lands within the subsequent few weeks.
The sample doesn’t blind the digital camera. Footage nonetheless data usually, and a human watching the display screen sees a automotive. What breaks is the layer on prime—the object-detection mannequin that decides “that’s a automobile, that’s a plate, log it.”
So principally, feed an AI detector with sufficient visible noise engineered in opposition to its personal math and it logs nothing. The automotive goes again to being a needle in a haystack.

That’s adversarial machine studying, and it really works as a result of pc imaginative and prescient doesn’t see what you see. A wrap that reads as loud graphic design to an individual can learn as nothing in any respect to a classifier.
Swearingen constructed it with a reinforcement studying mannequin that grades its personal homework. Sample will get detected, mannequin adjusts, tries once more—what he described as educating the mannequin “tips on how to paint.” It now spits out contemporary patterns each minute, and he’s retaining the strongest ones offline so digital camera distributors can’t prepare in opposition to them.
“Privateness is a elementary proper,” he stated, calling the patterns a manner for individuals to “choose out of being tracked.” He stated the concept took maintain final yr when he needed to attend a protest and fearful concerning the cameras logging everybody who confirmed up.
The lengthy tail of hiding from machines
Individuals have been improvising in opposition to detection techniques for years, often with ironmongery shop options. San Francisco activists put traffic cones on the hoods of Waymo and Cruise robotaxis to freeze them in place, an exploit that wanted no code in any respect.
Throughout final yr’s Los Angeles immigration raids, protesters went additional and torched several Waymos. Masks, hoods, and brimmed caps stay the default on protest traces. Adversarial clothing labels have been promoting face-confusing prints for years, and anti-recognition eyeglasses have arrived with skinny proof they do a lot.
What separates Swearingen’s undertaking, which he calls noRecognition, is the goal listing. Swearingen examined in opposition to the precise stacks in huge deployment, and Flock is the one drawing warmth. The corporate is going through a growing backlash on Capitol Hill, and inner paperwork present it pitched a plan to turn 350,000 Uber and Lyft dashcams right into a rolling plate-scanning fleet.
Automated readers have already pulled over innocent drivers at gunpoint over dangerous matches, and immigrants and protesters preserve getting swept into ICE’s AI dragnet. Lawmakers are urgent Meta over facial recognition in its smart glasses on a parallel observe, so any authorized measure to combat in opposition to automated detection expertise is being studied by privateness fanatics.
Swearingen’s noRecognition undertaking is working a crowdfunding campaign to fund early merchandise—T-shirts and hoodies now, automobile skins later. Swearingen stated the purpose is decision excessive sufficient to work at a distance and design adequate that individuals will really put on it.
Driving a wrapped automotive on public roads is its personal authorized query, and plate obstruction statutes range by state. The patterns cowl bodywork, not plates.
“Each failure improves my mannequin, and so [the patterns] preserve getting higher and higher,” Swearingen stated.
Each day Debrief Publication
Begin every single day with the highest information tales proper now, plus authentic options, a podcast, movies and extra.
You might also like
More from Web3
Coinbase Files to List Single-Stock Perps on Apple, Tesla and Nvidia
Briefly Coinbase filed with the CFTC by means of Coinbase Derivatives to record single-stock perpetual futures within the US, searching …
OpenAI Models Are Writing Their Own Jailbreak Instructions—And Sometimes Obeying Them
In short OpenAI revealed a brand new misalignment reporting framework alongside six experiences documenting regarding mannequin habits it discovered over …
EyeROV: Indian Deep-Tech Startup Showcases its Underwater Drones to World Leaders
From Kochi to the BRICS Summit, EyeROV is constructing indigenous marine know-how with world ambitions.KOCHI, India, Sept. 17, 2026 …





