When AI Gets You Wrong #shorts

Our Take

Credit to creator kaynemcgladrey for surfacing the Robert Dillon case in a tight, sourced short. It's a textbook example of what happens when law enforcement treats an algorithm's confidence score as probable cause instead of a starting point for actual investigation. A 93% match sounds authoritative until you learn the input was a grainy photo of a computer screen, and that officers allegedly buried negative license plate hits and a distinctive scar that didn't match — details a human investigator should never have ignored.

This case matters far beyond facial recognition. It's the same pattern we track with ALPR networks like Flock Safety: a black-box system generates a lead, officers treat that lead as confirmation rather than a hypothesis, and due diligence gets skipped because the tech feels objective. The Monell liability angle in this lawsuit — arguing the county's FACES database itself is systemically defective — is worth watching, because it's the same legal theory that could eventually be aimed at agencies that lean on ALPR hits without corroboration. A database of 38.5 million faces, or a statewide network of plate-reading cameras, is only as trustworthy as the humans reviewing its output, and this case shows that review can be nonexistent.

If you want to see how dense this kind of automated surveillance infrastructure already is in your own community, check our camera map, and if you'd rather do something about it, our take-action page has concrete steps for pushing back at the local level.