Facial St. Louis: One Number Jailed Wrong Man 17 Months

Christopher Gatlin spent about seventeen months behind bars. The thing that first pointed police to him? A computer's guess, built from a blurry frame of surveillance video at a St. Louis train station. According to his lawsuit, that guess became a path straight to a jail cell.

If you've ever walked past a security camera,

If you've ever walked past a security camera, you've been in a frame just like that one. Most of us never think about what happens to those images. And it's unsettling to realize a grainy picture could put your name in front of a detective. According to reporting from the St. Louis Post-Dispatch, St. Louis County investigators ran a still image from a MetroLink camera through facial recognition software. MetroLink is the region's light rail system. The software suggested Gatlin. Now he's suing St. Louis County. According to Bloomberg Law, his complaint also targets Amazon's facial recognition tool, called Rekognition. So how does one computer suggestion turn into more than a year of someone's life?

Start with the words officers use. According to Washington Post reporting, some officers describe unverified matches as a "one hundred percent match." That's not what these systems produce. They produce a ranking. A list of faces that look similar, with a score attached. That score is a probability, not an identity. And the photo feeding it matters. A dark, blurry frame from a train platform gives the software very little to work with. For investigators, that's the difference between a lead and a conclusion. For everyone else, it means a machine's best guess about your face can sound, out loud, like a certainty.

Then there's training. Gatlin's lawsuit alleges St. Louis County handed officers this technology without adequate training. And a judge found problems with what came next. In February of twenty twenty-four, Judge Brian May threw out the witness identification in the case. He found officers hadn't followed accepted procedures for a fair lineup. A fair lineup is supposed to keep investigators from nudging a witness toward a particular face. Once the software offered a name, that name became an anchor. Everything after it bent toward Gatlin, instead of testing whether he belonged there at all. Most of us would trust a witness who says "that's him." But what if the witness was quietly steered before saying it?

Gatlin isn't alone. According to the Washington Post, at least eight Americans have been wrongfully arrested after investigators leaned on facial recognition and skipped verification steps. The Innocence Project counts at least seven confirmed misidentification cases. Six of those people are Black. That pattern tracks with federal research. The National Institute of Standards and Technology, known as N.I.S.T., found some algorithms produced false matches up to a hundred times more often for Black and Asian faces than for white men. A hundred times. Estimates put popular systems right roughly nine times out of ten. That sounds reassuring. Until you remember the tenth result is a real person, with a family and a job and a front door.