A wrongful arrest from facial recognition is reported

The New York Times, working with the ACLU of Michigan, reported the case of Robert Williams, a Black man arrested by Detroit police on his driveway in front of his wife and young daughters on 9 January 2020, on suspicion of stealing watches from a Shinola store nearly two years earlier. The arrest followed a facial-recognition search: officers had submitted a blurry still from the store’s surveillance footage to the Michigan State Police, whose system — built by DataWorks Plus using Michigan’s driver’s licence database — returned Williams’ expired licence photo as a match. Williams was held for roughly 30 hours before an interrogating officer, holding the surveillance still next to his face, conceded “the computer must have gotten it wrong.” Prosecutors dismissed the case on 23 January without explanation to him.

The ACLU of Michigan filed a complaint against the Detroit Police Department the day the story ran, arguing the department’s reliance on the match, without adequate corroborating evidence, violated Williams’ rights; it separately sued federal agencies including the FBI and ICE seeking disclosure of their own facial-recognition use. Williams was reported as the first person publicly documented to have been wrongfully arrested in the United States on the basis of a facial-recognition match, though researchers including Joy Buolamwini had already shown the technology performed markedly worse on darker-skinned faces, and defence attorneys and reporters have since suggested other, unreported cases likely preceded it.

The case surfaced two weeks after IBM, Amazon and Microsoft announced restrictions on police sales of facial-recognition systems, giving campaigners a concrete example to set against the technology’s error rates. Williams later sued the city; a 2024 settlement required Detroit to overhaul how its police department could use facial recognition, including barring arrests based on a match alone.