AI Pattern on a Car Confused Police Software. Dozens of American Cities Cancel Surveillance Networks, Slovakia Resolves the Origin of Cameras

At the DEF CON hacker conference in Las Vegas, a Toyota Yaris covered with a computer-generated pattern passed by a camera of the Flock police surveillance network and its software did not recognize the vehicle as a car. Italian Corriere della Sera reported on it Saturday morning in the Motori section as a technical curiosity, while Estonian Postimees that same morning framed it as a "war with spy cameras" that Americans are winning at city council meetings. Of Slovak media, only Pravda covered the topic, but wrote only about the print on a t-shirt — not about the car and not about the network that the pattern confused.

The Same Hacker, Two Different Stories

Behind the pattern stands Bill Swearingen, a security researcher from Kansas City and co-founder of the SecKC community meetup. His noRecognition project generates graphics using reinforcement learning — according to TechCrunch, the model "learns to paint" and has gone through approximately 31 million tests over a year. TechCrunch ↗

The pattern doesn't work by covering a face, license plate, or car body from the camera. The camera still captures the person and car normally, but the analytical software analyzing the recording cannot achieve sufficient certainty that it is a person or vehicle, and does not trigger a detection alert. Swearingen claims his patterns fooled all eleven detection algorithms he tested — these were publicly available models, but according to him, they include those used by Flock's license plate readers, Axon's body-worn cameras, and the Clearview AI system. These are his own tests conducted independently, which no one has yet independently repeated, and according to his own words, he does not publish his strongest patterns to prevent camera manufacturers from preparing countermeasures. TechCrunch ↗

What does it look like? On clothing, the pattern resembles high-contrast black-and-white concentric circles that look like targets. The print that Swearingen applied to the Yaris was different and wilder — the Carscoops website compared it to a 1990s jet ski catalog. What they have in common is that to the human eye they are just striking graphic motifs, while for software they are deliberately prepared interference signals. Carscoops ↗

Ako to funguje

Vzor nezakrýva obraz — mieri na krok, v ktorom softvér rozhoduje, či na zázname vôbec je vozidlo.

Corriere made it a story about a race between detection models and countermeasures — in its telling, a chess match in which one artificial intelligence stands guard and another looks for a hole. It emphasizes that this is not an urban legend about reflective film that makes a license plate unreadable. The pattern is applied to the car body and attacks how software interprets shape, not the readability of the plate. And it adds to the scale of the problem: the Flock network, according to the Italian newspaper, captures approximately 20 billion license plates monthly with accuracy ranging from 93 to 99 percent. This is data from company materials, but even so, at such a volume, even an excellent-sounding percentage produces enormous absolute numbers of possible misreadings. Corriere della Sera ↗

Postimees structures the text oppositely from the publicly available opening. It does not begin with the hacker, but with the claim that automatic license plate readers are disappearing from American streets — not because cities no longer need them, but because resistance to mass surveillance is too loud to ignore. Flock's cameras are being sabotaged and municipalities are terminating contracts. The pattern on clothing in the Estonian account is only the final step of those for whom that is not enough. Postimees ↗

What Is Missing from Slovak Coverage

The newspaper Pravda reported on Swearingen's project on Thursday, correctly and with figures from tests against the YOLOv5 models. However, it remained focused on the print on clothing and on the technical question of whether the pattern works. Pravda ↗

What is missing is the crucial part that both foreign editorial offices have: the political dimension. According to the DeFlock organization, which keeps records of canceled and suspended contracts, over fifty American government agencies and municipalities have canceled, suspended, or declined contracts with Flock since January 2026. The company responded to the pressure this week with a package of changes that authorities must implement no later than January 1, 2027 — shortening standard data retention from thirty to seven days, requiring a case number before each search, automatic account lockout at suspicious search patterns, and the ability to block queries from other agencies on selected crime types. The American Civil Liberties Union called shorter retention a step in the right direction, the rest insufficient. Lawyer Robert Frommer from the Institute for Justice speaks of cosmetic improvements by the company in "panic mode" and insists that access to records should be decided by a judge, not a police officer. Boston.com ↗

Slovak Debate Running on a Parallel Track of the Same Railway

The coincidence is striking. The same week Americans are deciding who gets to look through their cameras, the National Security Authority warned that traffic radar cameras purchased for the Interior Ministry are a significant cybersecurity threat — due to the origin of hardware and software, remote access methods, and weak security. Aktuality.sk ↗

Read also: [NSA Warns of Russian Components in Traffic Cameras](/article/nsa-warns-of-russian-components-in-traffic-cameras-0814-1201)

It is not the same case or the same hardware. The American Flock network reads license plates and shares data between police departments, Slovak devices are speed cameras and the problem is their origin and security. What unites them is the same blind spot: a traffic camera is not a lens on a pole, but a network device with software, access rights, and someone who updates it. The American question is: who has the right to look. The Slovak question is: who is already looking without the operator knowing it. Denník N ↗

Read also: [Cameras of British Naval Drones Were Sending Data to China](/article/cameras-of-british-naval-drones-were-sending-data-to-china-1336547)

Limits of This Perspective

Corriere wrote from the Motori section, which is visible in the text — technology and fascination with cars outweigh the legal dimension. Postimees has the text behind a paywall, so we work from its public opening, and Estonian surveillance debate is naturally colored by proximity to Russia.

A more significant objection concerns the research itself. The effectiveness of the patterns so far rests on Swearingen's own tests, which TechCrunch reported — not on independently verified results. The promised demonstration video is not yet public and no one has repeated the attempt. Moreover, testing was conducted against publicly available models and it is uncertain whether Flock or Axon use exactly these same models in the field today. Swearingen himself admits that car wheels remained recognizable to the software and that the pattern requires sufficient print quality and resolution to work even from a greater distance. The distinction between car body and license plate is not universal either — it makes sense in the American legal environment and says nothing about how Slovak authorities would view such a modified car.

What hotinfo is Following