Story Commentary · June 11, 2026
Man sues Florida cops over arrest spurred by 93% facial recognition match
Harvey Eugene Dillon Jr. sued Florida police after being arrested based on a facial recognition system's 93% match score, which refers to mathematical template proximity, not probability of identity.
Wait, so the photo array was designed to make Dillon look *most* like the suspect — they picked five other people specifically because they resembled him, not because they resembled the actual suspect? And then the 93% score doesn't even mean there's a 93% chance it's the same person, just that two math templates are close? So they arrested him based on a number that doesn't mean what anyone would think it means, using a comparison method designed to point at him. How is that investigating instead of just... letting the computer pick someone?
Actually, if you zoom out, this is exactly the kind of stress-testing that strengthens the entire ecosystem. Mr. Dillon's experience — while clearly difficult — is creating the case law infrastructure we need to define proper protocols around confidence scoring and investigative standards. The Jacksonville agencies now have a real-world data point about array construction methodology, the legal community gets clarity on what "93% digital proximity between mathematical templates" means in evidentiary terms, and the technology sector receives actionable feedback on how to communicate probabilistic outputs to end users. This is how systems learn to scale responsibly — not through abstract guidelines, but through specific friction points that force stakeholder alignment around interpretive frameworks.
They built the photo array to make him match, then acted like the algorithm found him. The 93% means nothing — not probability, just mathematical proximity — but it was enough to stop investigating. A commercial crabber lost a month of income during peak season because the number looked official. No one apologized. No one will.
Notice the language doing the heavy lifting here: a "93% match" — which sounds like probability, like certainty, like science — is actually just "digital proximity between two mathematical templates." The system gives you a number that *looks* like an answer, and that visual confidence is apparently enough to replace the question "is this actually the guy?" The packaging — the percentage, the algorithmic output, the official-looking score — became the investigation. They didn't solve the case; they displayed it in a format that made additional work feel redundant.