The Sticker That Blinds the AI

If you watch one thing to understand Flock, make it this. Technologist and musician Benn Jordan — the same researcher who helped expose Flock cameras left open on the public internet — made a roughly 3-million-view video, “Breaking The Creepy AI in Police Cameras,” that is the clearest, most entertaining breakdown of ALPR surveillance we’ve seen.

What makes it brilliant

Jordan walks through how the cameras read plates — image segmentation, then OCR — and then does something genuinely clever: he demonstrates adversarial machine learning. He trains a model to generate “adversarial noise,” a subtle pattern that, applied to a plate, causes ALPR neural networks (like YOLOv8 or OpenALPR) to misread it or fail to detect it entirely — while a human’s eye still reads the plate perfectly. The lesson lands hard: these systems aren’t infallible oracles. They’re brittle pattern-matchers that can be confused by design. (The Decoder; Hackaday.)

He open-sourced the technique as PlateShapez — a Python research tool (MIT-licensed, 500+ stars) that automates “small, hardly-noticeable, random gaussian shapes to confuse AI license plate readers used by law enforcement and retailers.” Tellingly, the repo ships with its own ethics guidance and a dataset card, and frames itself as research into adversarial robustness. That’s exactly the right way to hold it: a window into how fragile these systems are — not a kit for your car.

He also documents the rot we cover in our teardown: unpatched Android, credentials on the dark web, exposed feeds. As public education, it’s superb — share it widely.

Now the part we have to say plainly

Do not put an adversarial sticker on your license plate in North Carolina. However elegant the science, applying anything to your plate that defeats its readability is almost certainly illegal here. North Carolina law requires your registration plate to be legible and unobstructed (see N.C. Gen. Stat. § 20-63), and altering or covering a plate to thwart reading invites a citation — and potentially worse if it’s treated as obscuring or altering the plate. A transparent pattern that makes your plate scan as the wrong number is exactly the kind of alteration that law is built to catch.

The value of Jordan’s work is understanding, not evasion. You don’t beat mass surveillance one bumper at a time — a single modified plate just makes you the interesting car. The moves that are both legal and effective are collective: detect the cameras, map them, and show up where the contracts get decided.

This article is general information about North Carolina law, not legal advice, and does not create an attorney-client relationship. If you are involved in litigation, some steps described here may conflict with a duty to preserve evidence or an existing protective order. Consult a licensed North Carolina attorney before acting.