Robert Williams is arrested after a false facial-recognition lead

In January 2020, Detroit police arrested Robert Williams at his Farmington Hills home in front of his wife and young daughters. He was accused of stealing watches from a Detroit store in 2018. He spent about 30 hours in custody before his release; the charges were later dropped. Williams and his lawyers say he was nowhere near the theft. The arrest and its effect on his family are central to the case, alongside the technology that helped put his name before investigators.^[1][3]

How the identification became an arrest

According to the ACLU’s case account, investigators sent a poor-quality surveillance still to Michigan State Police for a facial-recognition search. A possible match to an old driver’s-license photo of Williams led police to put his photo in a lineup. The person who selected it was an off-site loss-prevention contractor who had not witnessed the theft and had seen the same surveillance footage. The ACLU alleges that the warrant application left out information bearing on the reliability of both steps. These are the plaintiff’s account and allegations, not findings after a trial.^[1]

The distinction matters: a candidate returned by a search is an investigative lead, while a lineup selection based on the same image does not necessarily supply independent corroboration. NIST’s demographic evaluation documents performance differences across algorithms and populations; it did not test the system used in this investigation or determine why Williams was selected.^[4]

The case and settlement

Williams sued the city and officers in April 2021. A federal court entered the parties’ settlement on June 28, 2024. Its attached Detroit Police directive requires independent, reliable evidence linking a person to a crime before an arrest; a facial-recognition lead and a lineup identification alone are insufficient. It also bars a lineup based solely on a facial-recognition lead without an independent, reliable basis linking the person to the crime. The agreement provides for training and an audit of past cases, with court enforcement for four years. It expressly denies that the settlement is an admission of liability. The agreement states requirements; it does not by itself show how consistently they have been carried out.^[1][2]

In a Michigan Public interview, Williams welcomed the safeguards but said he had wanted Detroit police to stop using facial recognition altogether. He described the family trauma and the possibility that a more serious accusation could have had worse consequences. Detroit police told the outlet that their policy had changed after his arrest and that a match should serve only as an investigative tip. Those are the participant’s perspective and the department’s account, respectively.^[3]

In April 2026, BridgeDetroit reported that a Detroit Police spokesperson said the required audit was complete but could not immediately provide a summary of its findings. That statement reports completion; it does not establish what the audit found or whether the safeguards prevented further wrongful arrests.^[5]

Meaningful Human Control asks what a reviewer must actually check before acting on an automated lead. The incidents guide places this arrest alongside other deployment failures without treating one case as a measure of every facial-recognition system.