93% AI match was wrong and the real criminal escapes

Our Take

Unreasonable Doubts lays out a case that should terrify anyone who thinks a percentage on a screen equals proof: Robert Dillon, arrested in Florida in 2024 after facial recognition software spit out a "93% match," while the person who actually committed the crime remained free. Credit to the channel for walking through how quickly a confidence score becomes a court case, and how little scrutiny that number gets once it's in an officer's hands.

This is the same rot at the core of Flock Safety's ALPR network. A camera flags a plate, an algorithm says "match," and a human being treats that output as ground truth instead of a starting point. Whether it's a face or a license plate, the pattern is identical: opaque software makes a probabilistic guess, law enforcement treats it as certainty, and the burden of proof quietly shifts onto whoever got flagged. Dillon's case is facial recognition, not plate-reader tech, but it's the same surveillance-industrial logic — trust the machine first, ask questions later, and hope the wrongly accused has the resources to fight back.

Every one of these systems runs on the assumption that citizens won't push back on the infrastructure itself. Find out if Flock cameras are already logging your plate on our map, and if you want to do something about it, our take-action page has concrete steps for confronting this stuff at the city council level, before it's your face or your plate generating the next false "match."