CaraComp
CaraComp
Forensic-Grade AI Face Recognition for:
Get Started7-day refund guarantee**
facial-recognitionBy Cara Candelario

Facial Recognition Systems: 26,000 NYPD Searches Unchecked

facial recognition, face match, a police analyst comparing two face photos on a screen
A police analyst compares two faces in a facial recognition search, a match that should be a lead, not proof. Illustration: CaraComp

Quick answer

Can facial recognition match the wrong person?

Yes. Facial recognition can return a stranger who merely looks similar, and the mistake stays hidden unless someone checks it. Error rates are higher for people of color, women and older adults. A match score is a measurement to weigh against other evidence, such as witnesses or location records, never proof.

Picture a computer telling a detective that your face looks like the person in a blurry store surveillance video. You weren't there. But now your name is on somebody's desk. That's the risk with facial recognition in a city of eight million faces, and the NYPD just ran about 26,000 searches with gaps in oversight.

TL;DR

Facial recognition can point police at the wrong person, so a face match should only be a lead that gets checked against other evidence, never treated as proof.

First, the plain answer to "am I in danger?" For most people, on most days, no. But the way face recognition gets used is what decides whether a wrong match stays a harmless mistake or turns into a very bad week. Let's walk through it.

What the NYPD facial recognition numbers say about face recognition oversight

The reporting comes from Patch. In 2025, the NYPD's Intelligence Division ran roughly 26,000 facial recognition analyses. About a third of them went ahead without a supervisor reviewing them. Think of it as 26,000 times someone typed a face into a machine and asked, "Who is this?"

26,000
NYPD facial recognition searches in 2025, about a third without supervisor review
Source: Patch

There's more. Investigators reportedly used publicly available photos of high-profile people as stand-in images for searches. That breaks department policy. It also tells you something about the culture: this was treated like a casual tool, not a careful one. The city's Department of Investigation looked into it and made 17 recommendations, as ID Tech Wire reports.

One honest caveat: the headline doesn't name a person who was harmed by these searches. I'm not going to invent one. The point here is the setup, not a single victim. This article is part of a series, start with Fargo Facial Face Match Jails Tennessee Woman 108 Days.

Why a facial recognition match is only a lead

A match score (a number for how alike two faces are) is a guess, not a verdict. The NYPD's own policy says a match is "solely an investigative lead and not probable cause for arrest." Good rule. The trouble is what happens when nobody checks that it's followed.


Can facial recognition technology identify the wrong person?

Yes. Facial recognition can return a face that looks similar but belongs to a stranger. Studies show it is less accurate for people of color, women, and older adults. The Innocence Project counts at least seven confirmed misidentification cases, six involving Black people.

Face recognition technology uses complex algorithms built on computer vision and machine learning. Facial detection first finds a person's face in an image, such as a still from a security camera. Then the software identifies key facial landmarks, measures facial features such as the distance between the eyes, and turns them into data. Facial recognition systems compare that data against a database of images and return the closest candidates. They cannot identify anyone with certainty.

Here's the part that sticks with me. Once a computer names someone, humans tend to believe it. That's confirmation bias (we look for proof of what we already believe). A detective builds a photo lineup around the computer's pick. The witness sees a familiar face. A warrant follows. The ACLU explains why a simple warning label doesn't stop this. The Robert Williams wrongful arrest case, covered by Lexipol, is the textbook example.

any facial recognition match is solely an investigative lead and not probable cause for arrest

NYPD policy, as reported by Patch

What does a wrong match look like in real life?

A wrong match means a similar-looking stranger gets flagged, and the error is invisible unless someone checks it. Researchers also found East African women faced roughly 100 times more false positives (wrong matches) than white men. That gap is what makes a single, unchecked search risky.

Why NYPD facial recognition oversight matters

  • Scale26,000 searches means even a small error rate touches real people.
  • No second pair of eyesa third of searches skipped supervisor review.
  • Missing recordwithout an audit trail (the record of who looked at what, and when), nobody can retrace a bad match.

Trusted by Investigators Worldwide
Run Forensic-Grade Comparisons in Seconds
Detailed facial comparison reports. Results in seconds.
Get Started
7-day refund guarantee**

NYPD facial recognition systems: a lead versus proof

Here's the contrast that matters. The same match can be handled two very different ways. Previously in this series: Anti Facial Recognition Makeup Built To Fool 2001 Cameras.

StepCareful useCareless use
What the match isA lead to checkTreated as an answer
Who reviews itA supervisor signs offNo review
What backs it upOther evidence, like location or witnessesA lineup built around the match
Record keptDocumented audit trailNothing to retrace

Defenders have a fair point. In Florida, facial recognition helped clear a man falsely accused of vehicular homicide by leading to a key witness. The technology isn't the villain. Unchecked use is. As the Columbia Human Rights Law Review argues, people accused also have a right to know about false-match risk.

If you've ever wondered whether a photo or a face is really who it claims to be, that's the exact question facial recognition technology exists to answer. Here's the one thing worth watching for: ask whether anyone compared the faces side by side and wrote down how. A careful comparison leaves a paper trail. A hunch doesn't.

Amnesty International has also documented policy violations, including a case that involved an Instagram photo and one involving a music video.

Key Takeaway

Facial recognition is a lead generator, and NYPD facial recognition use only stays fair when a human checks the match against other evidence and writes down how.

Recap: A face match can point police at the wrong person. It should be checked against other evidence before anyone is treated as guilty.

So, one more time: if a face match pointed police toward you, would you expect to be told how it was checked? Right now, with a third of 26,000 searches unreviewed, nobody can promise you that answer. Up next: Facial Recognition 26 000 Nypd Searches A Third Unchecked.

Facial recognition technology: Frequently Asked Questions

Is a facial recognition match enough to arrest someone?

No. The NYPD's own policy says a match is only an investigative lead, not probable cause (the legal standard for an arrest). A match should be checked against other evidence, such as location records, witnesses, or physical proof. Cases of wrongful arrest happened when police skipped those checks and trusted the computer's pick. A face match is a starting point for work, not the end of it.

How many searches did NYPD facial recognition run?

The NYPD's Intelligence Division conducted about 26,000 facial recognition analyses in 2025, according to Patch. Around a third of them proceeded without supervisory review. That means no second person looked at the search before it was used. The number matters because even a small error rate, applied to that many searches, can mean real people get flagged by mistake.

Does facial recognition work equally well for everyone?

No. Studies show facial recognition is less accurate at identifying people of color, women, and older adults. One finding cited in the research showed East African women had roughly 100 times more false positives than white men. A false positive is when the recognition software says two different people are the same person. That is why human checking matters most for those groups.

What is an audit trail in a face search?

An audit trail is the record of who ran a search, what image they used, who reviewed it, and when. It lets someone retrace how a match was reached. Without it, a mistaken match is hard to catch and hard to challenge. Documented records protect both investigators doing honest work and people who might be wrongly accused.

Can police use a celebrity photo in a face search?

Not under NYPD policy. Patch reports that investigators used publicly available photos of high-profile people as substitutes for real probe images (the photo being searched), which violated department rules. It suggests searches were sometimes run casually rather than with careful investigative rigor, which is part of why critics want stronger review before any search is run.

Ready for forensic-grade facial comparison?

Full forensic reports with detailed similarity scoring. Results in seconds.

Run My First Search