Facial Recognition Wrongful Arrest: Police Trusted a 93% Match — It Was Wrong

Our Take

Ninety-three percent sounds like certainty. It isn't. As AI by Age lays out, that figure is a similarity score between two images, not a probability that Robert Dillon committed a crime. Police and prosecutors routinely collapse that distinction, and the people who pay for it are the ones staring down a felony charge because a machine said their face looked close enough to a suspect's.

This isn't an isolated glitch. Porcha Woodruff, eight months pregnant, was hauled in on a carjacking charge because Detroit PD trusted the same kind of output. These systems are trained and tested unevenly across race and gender, deployed with almost no public oversight, and treated by officers as a shortcut around actual investigative work. A 'match' becomes a warrant becomes an arrest, and the burden of proving innocence lands on the person who was never a suspect in the first place.

Flock Safety's ALPR network is a cousin to this problem — a growing web of cameras logging your movements with no meaningful audit trail, feeding data to whoever asks. If your town runs Flock cameras, check our map to see where, and use our action guide to push your local officials on policy before the next wrongful arrest is someone in your own community.