Florida Cops Arrested a Man on a 93% Facial-Recognition Match - He Was 300 Miles Away

A facial-recognition algorithm spit out a 93% match, and cops ran with it — ignoring the fact that the suspect lived 300 miles from the crime scene.

93% Match, Zero Investigation

Robert Dillon, 52, was arrested in August 2024 for allegedly trying to lure a child at a Jacksonville Beach McDonald's. The sole basis: a facial-recognition system that matched him to a low-quality photo — specifically, a picture of a McDonald's computer screen showing surveillance footage. The lawsuit, filed this week, says officers never tested that match against basic available evidence.

A search of license plate reader databases turned up zero hits for Dillon in Jacksonville Beach on the day of the incident. Dillon lives in Fort Myers, more than 300 miles away. The cops reportedly didn't bother to check alibis or confirm his whereabouts.

How Cops Confirmed a False Positive

“This case is about what happens when police let an error-prone artificial intelligence system stand in for an investigation,” the lawsuit states. Dillon's legal team argues officers built a case to confirm the algorithm's output rather than test it. The arrest subjected him to one of the most stigmatizing charges possible, and the suit alleges officers concealed exculpatory evidence that would have cleared him before trial ever started.

The 93% confidence score is a number that looks precise but is meaningless without context — especially when the input image is a grainy screenshot of a screen recording. Any engineer who has worked with recognition models knows that confidence scores degrade drastically with image quality, and a 93% match on a downsampled, recompressed still is not trustworthy evidence.

Policy Gap That Lands Innocent People in Cuffs

This isn't a story about a bad algorithm — it's about law enforcement that treated a black-box output as a truth signal. The lawsuit doesn't just target the facial-recognition vendor; it names the individual officers and the Jacksonville Beach Police Department for adopting the match without independent verification.

If courts allow this kind of faith-based policing to stand, every jurisdiction using facial-recognition systems will face similar liability. The question isn't whether the algorithm can be wrong — it's whether police have a duty to check before they arrest. Dillon's case says they do, and this suit aims to prove it.

Source: Man sues Florida cops over arrest spurred by "93% match" in facial recognition
Domain: arstechnica.com