The Terrifying Math of AI Surveillance: 1.4 Billion Errors Per Month? 👁️

Our Take

Larry Brock's latest episode of The Reckoning does something we appreciate: it takes the marketing language around 'AI-powered public safety' and runs the numbers. Flock Safety and its competitors love to tout accuracy percentages that sound impressive in a press release, but as Brock points out, even a system that's right 99.9% of the time generates enormous volumes of errors once you scale it up to the billions of plate reads these networks rack up nationwide every month. That's not a rounding error — that's potentially thousands of innocent drivers getting flagged, stopped, or investigated based on a database mistake.

This is the part of the surveillance debate that gets buried under vendor talking points. Nobody disputes that ALPR cameras can read plates fast. The question is what happens downstream when a misread, a stale hotlist entry, or a bad match sends an armed officer to pull over the wrong car. Brock frames it as a math problem, and he's right to — scale is exactly how 'rare' errors stop being rare. A system doesn't need to be malicious to cause harm; it just needs to be deployed at the volume Flock and friends are currently operating at, with little independent auditing of how often these errors actually occur or get corrected.

We'd add one thing Brock's video gestures at but can't fully cover in one episode: this isn't hypothetical anywhere near you. These cameras are likely already logging your plate on your daily commute, feeding a database you have no access to and no way to challenge. Check our camera map to see what's already watching your neighborhood, and head to our take-action page if you want to push back at the local level.