Facial Recognition Gets People Arrested by Mistake: Why It's a Bigger Problem Than You'd Think

Police in the United States have arrested at least fourteen people who were later proven innocent, based on misidentifications made by facial recognition technology. An ACLU report from April 2026 documented these cases. One woman, identified only as Ms. Williams, spent six months in jail before police realized they had the wrong person. She is now asking the Maryland police departments involved for a formal apology.

This is not a rare accident. It is what happens when police treat a tool that makes educated guesses like it makes certain identifications.

How Facial Recognition Actually Works — and Where It Fails

Facial recognition does not say "this is the person." It says "this person's face looks most similar to the person in this photo." The system examines a picture — often a blurry or poorly lit frame from a security camera — and returns a ranked list of possible matches from a database. The top result looks most like the photo, but that does not mean it is the right person. It just means the computer found the closest match it could.

The ACLU has pointed out that the top match is often wrong. The system is making a statistical guess, not confirming an identity.

The U.S. Commission on Civil Rights said this clearly in a September 2024 report: facial recognition results alone cannot prove someone committed a crime. They are a starting point for investigation, not proof of guilt.

The accuracy is not the same for everyone. Research has found that facial recognition systems make more mistakes when identifying people with darker skin, women, and older adults. This becomes especially dangerous when police use it to decide whether to arrest someone.

Cases Where People Were Wrongly Arrested

Detroit has had the most documented cases — at least three people arrested based on facial recognition errors according to the ACLU. In New York, a man spent two days in jail for a crime he did not commit after an NYPD facial recognition error in August 2025. A University of Michigan Law School study examined another case where a suspect was arrested based almost entirely on what the facial recognition system suggested.

What connects these cases is the same problem: once the facial recognition system picked a name, police treated that as a solved case instead of as a lead to investigate further. They did not look hard enough for other evidence — witnesses who could confirm identity, proof of where the person actually was, phone location records. When they did collect other evidence, they did not give it enough weight.

The Legal Risk Is Real

Police departments and cities can face lawsuits for wrongful arrests based on facial recognition errors. These lawsuits can claim that police violated the Fourth Amendment (rules against unreasonable arrest) or the Fourteenth Amendment (rules about fair treatment). Private companies can also face legal trouble. The Federal Trade Commission banned Rite Aid from using facial recognition in 2023, finding that the pharmacy chain had used the technology poorly and flagged innocent shoppers, mostly in lower-income and minority neighborhoods. Back in January 2020, the Ohio Attorney General warned that private companies could face serious liability if they used facial recognition to catch shoplifters but got it wrong.

Some Police Departments Are Making Changes — But Not All

Detroit eventually created what the ACLU called the nation's strongest police policy on facial recognition in mid-2024. That policy says that facial recognition must not lead to an arrest without other evidence to back it up. That is a good safeguard, but it took multiple wrongful arrests in one city to make it happen.

The federal government has not passed any law that sets rules for how police across the country must use facial recognition. Different cities and states have different rules. Most police departments do not have to prove a suspect did something wrong before making an arrest based on facial recognition.

The public count of fourteen wrongful arrests almost certainly leaves out many cases. Only arrests that end up in civil court or get news coverage make it into public record. Cases that are quietly dismissed — especially if the person arrested cannot afford a lawyer — often stay hidden. The real number of innocent people arrested based on facial recognition is probably higher than fourteen.

Why This Pattern Feels Familiar

This problem is not new. In the 1990s and 2000s, police started using automated fingerprint identification systems — computers that could quickly search millions of fingerprints. The system was powerful and useful, but it also created a false sense of certainty. When an examiner saw that the computer had ranked a fingerprint as a top match, they would treat it like proof, not like a clue. The Brandon Mayfield case in 2004 became the turning point — the FBI arrested an innocent Oregon lawyer based on a fingerprint match to a bombing suspect that the computer had suggested but that was ultimately wrong. Facial recognition is following the same path, but it is happening faster and affecting more people.

The real issue is not with the technology itself. It is about how people use it. The facial recognition system gives investigators a possible suspect. The investigator then needs to ask: "Is this person actually guilty?" They need to treat the computer's answer as a theory to test, not as a fact already proven. But police departments have not been reliable about doing this.

What Would Actually Fix This

The technical side is straightforward. Computer vision experts know what to do: require better quality images before running facial recognition, make the system tell investigators how confident it is in its answer, test the system fairly across all skin tones and ages, and keep detailed records of every search and what happened next. These fixes are possible right now, without waiting for better technology.

The procedural side is equally clear. Police departments need to require independent evidence — witness statements, location records, surveillance video from other sources — before making an arrest. These requirements need to be written in formal policy, not just mentioned in a training session. There also need to be real consequences when departments break these rules. Law enforcement agencies should regularly review their records to spot patterns of errors before more innocent people are arrested.

The core capability of facial recognition is not in question. It works well for specific purposes, when used the right way. The problem is using it beyond those boundaries, without the proper safeguards. Fixing this does not mean getting rid of the technology. It means handling it with the same care that forensic science — at its best — has always required.