Why Facial Recognition Keeps Getting People Arrested for Crimes They Didn't Commit

At least fourteen people in the United States have been wrongfully arrested because police relied on facial recognition software that made a mistake, according to an ACLU report from April 2026. One woman, identified only as Ms. Williams, spent six months in jail after being incorrectly identified. She is now demanding a formal apology from the Maryland police departments that arrested her.

These are not rare glitches in a system that mostly works. They are predictable consequences of how the technology is actually being used: as a definitive answer when it is really just a guess.

What Facial Recognition Actually Does

When police run a photo through facial recognition software — say, a blurry security camera image from a convenience store — the system does not return a match. It returns a ranked list of similar faces from a database, arranged by how closely each one resembles the probe image. The top result is simply the closest statistical match the algorithm could find. It is not confirmation that the person in the probe image is actually in the database at all.

This is an important distinction. The U.S. Commission on Civil Rights stated explicitly in a September 2024 report that facial recognition results alone do not count as positive identification and do not establish probable cause for arrest. This is not a fringe civil liberties claim — it is the considered conclusion of a federal agency after examining how police departments across the country actually use the technology.

The error rates also vary significantly depending on who is in the photo. Research has shown that commercial facial recognition systems are less accurate on darker-skinned individuals, women, and older adults. When such a system is used to help solve crimes, this disparity can have serious consequences.

How It Happens: The Cases

Detroit has become a center for documented wrongful arrests tied to facial recognition, with at least three confirmed cases traced back to Detroit police use of the technology. In New York, an NYPD facial recognition error led to a man being jailed for two days for a crime he did not commit, as reported in August 2025. A University of Michigan Law School analysis examined a case in which police relied almost entirely on a facial recognition result to identify a suspect — a decision that researchers described as both a wrongful arrest and a significant legal turning point.

The pattern in these cases is consistent: the police department treats the software's top result as conclusive proof rather than as a lead that needs to be investigated further. Background checks, witness statements, alibi information, and location data are either never gathered or are given less weight than the algorithmic result. By the time the arrest is made, the system's initial suggestion has hardened into assumed fact.

The Legal and Regulatory Consequences

Police departments that make arrests based on facial recognition errors can face civil lawsuits grounded in violation of the Fourth Amendment (unreasonable search and seizure) and due process rights, according to ACLU legal analysis. Cities and counties are liable both for the decision to deploy the technology and for training (or lack thereof) on how to interpret results correctly.

Private companies face regulatory pressure as well. The FTC issued a five-year ban against Rite Aid in December 2023 for deploying facial recognition in stores without adequate safeguards. The agency found that Rite Aid had flagged innocent customers as potential shoplifters, with a disproportionate impact in lower-income and minority neighborhoods. The Ohio Attorney General's Facial Recognition Task Force had warned as early as January 2020 that retailers using the technology to catch shoplifters could face serious liability if they arrested the wrong person — a warning Rite Aid's regulatory history bears out.

The Policy Patchwork

Detroit, after experiencing multiple wrongful arrests, eventually put in place what the ACLU called the nation's strongest police policy on facial recognition as of mid-2024. That policy restricts how police can use the technology as a lead and requires independent verification before any arrest. It is a meaningful safeguard, and it shows what good policy looks like. But it took several documented wrongful arrests in one city to force the change.

The federal government has not created a comprehensive set of rules governing how police use facial recognition as of June 2026. Instead, there is a fragmented landscape of city bans, state restrictions, and individual police department policies. Most jurisdictions still have no binding requirement that police corroborate a facial recognition result before making an arrest.

The actual number of wrongful arrests caused by facial recognition is almost certainly higher than fourteen. The ACLU count reflects only the cases that become public through civil lawsuits or news coverage. Cases that are quietly dismissed, or in which the arrested person lacks legal representation to fight back, rarely surface in the public record. How many people have been wrongfully arrested but never made news is unknowable.

A Pattern We Have Seen Before

Journalists who covered law enforcement technology in the 1990s and 2000s may find this arc uncomfortably familiar. Automated fingerprint identification systems (AFIS) were transformative tools — powerful, widely praised, and genuinely useful — and they also produced wrongful convictions when fingerprint examiners treated probabilistic similarity scores as certainties. The Brandon Mayfield case in 2004 became a landmark example: the FBI misidentified an Oregon attorney as a suspect in the Madrid train bombings based partly on a faulty fingerprint match. Facial recognition is following the same script, but faster and on a much larger scale.

The root problem is not actually a flaw in the algorithm itself. It is a failure in how the results are used. The technology produces a candidate — a lead worth investigating. The human investigator then needs to treat that candidate as a hypothesis to be tested, not as a conclusion ready for arrest. Across police departments, that distinction has not been reliably maintained.

What Would Actually Fix This

Computer vision researchers and forensic experts have identified the technical improvements: higher standards for the quality of images submitted to the system, uncertainty scores attached to every result (so officers can see how confident the algorithm is), testing the algorithm on diverse faces as part of procurement, and detailed logs that record every query and what happens as a result. None of these require waiting for better technology.

The policy and operational changes are equally straightforward: police departments must independently verify facial recognition results before making an arrest, and this requirement needs to be written into formal policy, not just offered as informal guidance; cities and departments need to face legal liability when they fail to follow proper procedures; and departments should regularly check their query logs against arrest records to spot systemic patterns of error before they result in a fifteenth, twentieth, or fortieth wrongful arrest.

Facial recognition does work for legitimate applications, within its actual limits. The problem is using it beyond those limits, without the disciplined procedures needed to manage what can go wrong. Fixing this gap does not mean abandoning the technology. It means holding it to the same standards of evidence that responsible forensic science has always required.