Facial Recognition: 26,000 NYPD Searches, a Third Unchecked
Quick answer
Why is unchecked facial recognition by police a problem?
A facial recognition match is only a lead, not proof, so skipping human review risks turning a software guess into an arrest. Once a computer names someone, investigators tend to look for confirmation. Error rates also differ across groups, which makes a supervisor's check and documented verification of every match essential.
Facial Recognition: 26,000 NYPD Searches, a Third Unchecked
This episode is based on our article:
Read the full article →Facial Recognition: 26,000 NYPD Searches, a Third Unchecked
Full Episode Transcript
Police in New York City ran about twenty-six thousand facial recognition searches last year. According to the city's own review, about a third of them happened without a supervisor checking the work. And in some searches, investigators didn't even use a photo from a crime scene.
If you've ever had your picture taken in public,
If you've ever had your picture taken in public, this story touches you. New York has about eight million people, and every one of those faces can end up compared against a database. Most of us never signed up for that, and most of us will never know if it happened. The report comes from the city's Department of Investigation, which reviewed how the N.Y.P.D. Intelligence Division used the technology. It found a pattern of searches with little review and policy violations along the way. So what happens when a computer's guess goes unchecked, in a city this size?
Start with the violation that's easiest to picture. According to the report, investigators used publicly available photos of high-profile people as stand-ins for the photo they were supposed to search with. That's against department policy. It suggests a tool being used casually, like a search bar, instead of a forensic instrument. Amnesty International has pointed to similar examples in its reporting, including one where police used an Instagram photo and another involving a music video. A tool this powerful, treated this loosely.
Now the part that's supposed to be the safeguard. The department's own policy says a facial recognition match is only an investigative lead. It is not probable cause for an arrest. In plain English, a match is a hint. It's not proof. But a hint with no supervisor looking at it has a way of hardening into a fact. The A.C.L.U. has argued that a simple warning on a screen doesn't stop that from happening. In wrongful arrest cases, police got warrants after a false match, then built a photo lineup around the face the computer picked. Once the computer names someone, everyone after that tends to look for confirmation. That's what experts call confirmation bias. For detectives, it's a process problem. For the rest of us, it means a software guess can quietly steer a human investigation.
The guesses aren't equally reliable for everyone
And the guesses aren't equally reliable for everyone. According to research cited in the Bulletin of the Atomic Scientists, drawing on federal N.I.S.T. testing, these systems make more mistakes on people of color, women, and older adults. East African women saw roughly one hundred times more false matches than white men. One hundred times. The Innocence Project counts at least seven people misidentified by this technology. Six of them are Black. Robert Williams, arrested in Detroit, is one of them.
There's also a history here. The reporting says the N.Y.P.D. stopped tracking How To Bypass Age Verification 1 In 3 Teens Already Know accuracy back in twenty fifteen, after finding the error rate was too high. Then use of the tool grew anyway. Stop measuring the mistakes, and the mistakes stop showing up in your numbers.
Defenders have a fair point, and it deserves airtime. In Florida, facial recognition helped clear a man falsely accused of vehicular homicide by finding a key witness. The N.Y.P.D. says the tool has a legitimate purpose when it's used within limits. So the tool can help the innocent too.
The Bottom Line
That's the twist. The technology isn't the main problem. The missing check is. A match nobody reviews isn't a lead. It's a liability, for the person accused and for the investigator who trusted it.
Police ran tens of thousands of face searches, and a third had nobody double-checking. A match is only a hint, but without review, hints can turn into arrests. The fix isn't throwing out the tool. It's making sure a human verifies and documents every match before anything else happens. Whether you investigate cases or just walk down a city street, the question is the same: who checked the computer's answer? The full story's in the description if you want the deep dive.
Ready for forensic-grade facial comparison?
Full forensic reports with detailed similarity scoring. Results in seconds.
Run My First SearchMore Episodes
Anti Facial Recognition Makeup: Built to Fool 2001 Cameras
Researchers in one lab study put makeup on twenty volunteers, and cameras correctly identified them just one point two two percent of the time. That sounds like a privacy miracle. But almost none of i
PodcastFacial Recognition: Hong Kong Cameras Go Live, No Appeal
Hong Kong police say cameras already watching the city have helped solve nearly twenty-two hundred cases. Now the city plans to multiply those cameras twelve times over. And so far, officials haven't
PodcastFargo Facial: Face Match Jails Tennessee Woman 108 Days
A Tennessee woman spent one hundred and eight days in jail. Police in a city she says she'd never visited arrested her because a computer thought her face matched a photo on a fake ID. Another police de
