Facial Recognition Error Jailed a Grandmother; She Sues for $10M

Angela Lipps was in Tennessee when someone was cashing fake-ID checks at banks in Fargo, North Dakota. She says a facial recognition match told police she was the thief anyway. Then she lost months of her life to it.

Facial recognition should only ever hand police a lead to check, and Angela Lipps' $10M lawsuit shows what happens when a computer match gets treated like proof.

Facial Recognition Made Her a Suspect: What Happened to Angela Lipps

Lipps is a 50-year-old grandmother. In July 2025 she was arrested on bank theft charges tied to Fargo-area banks. According to ABC News, the trail began with surveillance footage of a woman using a fake ID. A facial recognition tool compared that footage against photos, and it landed on Lipps' social media picture.

She was held for months. She was released on Christmas Eve, after bank records showed she was in Tennessee when the crimes happened. Her lawsuit says she lost her home and her car along the way. Her complaint also says police knew a facial recognition match should not be the only reason to extradite and charge someone. (Extradite means to send a person from one state to another to face charges.)

Read that again. The proof she was innocent was sitting in bank records the whole time. Nobody looked until she had already paid for it.

Is this the same case as the Tennessee wrongful-jailing story?

No. It is a separate case. It follows the recent Tennessee lawsuit over an alleged false match, which is part of why this week feels like a pattern. Two stories do not prove a trend. But they should make anyone ask how many people never got the bank records that cleared them. This article is part of a series, start with Age Verification Windows 11 Sorts Kids Into 5 Age Boxes.

---

Why Can AI Facial Recognition Get It So Wrong?

AI facial recognition does not say "this is her." It gives law enforcement a list of possible candidates. That is a very different thing. The match score (a number for how alike two faces are) tells you who looks similar. It does not tell you who did it.

It helps to know the steps. Face detection finds human faces in a video frame. The software then measures facial features, such as the distance between the eyes, and turns them into a biometric template, which is a string of numbers. Facial recognition algorithms create that template from one face image and compare it with more other face images in a database. Face matching then returns the closest candidates, ranked by score. Every step depends on the quality of the facial images it is given.

Face detection and facial recognition are separate jobs. Face detection only answers whether a face is in the frame. Recognition decides whose face it is, and that second step is where a wrong match becomes a wrong suspect.

Quality matters enormously. An airport camera gets a clean, straight-on look at you. Police often get a grainy still from a ceiling camera. The Federation of American Scientists notes that controlled testing looks very different from real policing conditions. Surveillance images tend to be blurry, low-contrast, and taken from odd angles.

Error rates are also not the same for everyone. The government's testing lab, NIST, found that false positives (the system saying two different people are the same person) vary a lot across groups. As of March 2025, the top system produced 358 times as many false positives for West African women over 65 as for Eastern European men aged 35 to 50.

But isn't the technology very accurate?

Defenders have a fair point. NIST found that error rates in verification-type tests (checking one photo against one other photo) are generally below 1 percent, and many are below 0.5 percent. That is under good conditions. A grainy still compared against a social media photo is not good conditions. And a small error rate is cold comfort when you are the error.

International standards say a facial recognition result alone is not a sufficient basis for questioning, arrest, or detention. Summary of international principles, INCLO

---

What Is the Difference Between a Lead and Proof in Facial Recognition?

A lead is a reason to look closer. Proof is what survives the looking. Facial recognition produces leads. Good police practice, as described by Police1, requires a trained human examiner to confirm a candidate before anyone acts. The NYPD policy, for example, requires corroborating evidence first.

Here is the whole story in one table. Previously in this series: Ai Blackmail 1 Selfie And An Ai Model Is All Scammers Need.

Step	Lead (done right)	Shortcut (what is alleged)	Why it matters for facial recognition
Match result	A candidate to check	Treated as the answer	Recognition software ranks candidates; it does not confirm identity
Human review	Trained examiner confirms	Little or none	A person compares the facial features the software scored
Other evidence	Records, location, timeline	Not checked before arrest	Bank records cleared Lipps, and matching alone could not
Supervision	Oversight of officers' use	Complaint says none	Law enforcement needs clear rules for face matching

My blunt read: this is not mainly a technology failure. The lawsuit says Fargo police knew that weak supervision could lead to wrongful detention and did not supervise anyway. A tool that offers guesses will produce wrong guesses. The failure is what people did next.

---

What Can You Do About AI Facial Recognition Mistakes?

If you have ever wondered whether a photo or profile is really who it claims to be, that is the exact question this kind of comparison technology exists to answer. Used well, facial recognition is a careful first look, with a person checking the work.

Not every use carries the same risk. Banks and apps use facial recognition for identity checks, where the software verifies that a selfie matches an ID photo. Those systems often add liveness detection, which checks that a real, live person is in front of the camera and not a printed photo. That supports fraud detection, fraud prevention and security. Searching a crowd or a blurry video frame for an unknown person is a much weaker kind of facial recognition, because nothing confirms who is in the image.

Phone unlocking and app login are a different case again. That kind of authentication compares you with biometric data you enrolled yourself, so the question is whether it is you, not who you might be. Many phones keep that data on the device rather than in the cloud, which helps keep it secure and protects your privacy.

Here is the one thing to watch for. When anyone says "the system matched this face," ask what else backs it up. A location record, a timestamp, a witness. If the only answer is the match itself, that is a lead dressed up as a fact. Lipps' alibi was a bank record. It existed from day one.

So here is the question I cannot shake. Lipps was cleared by something as ordinary as a bank statement. How many people had the same statement, and nobody ever asked to see it? Up next: Facial Recognition Grandmother Jailed Months Sues For 10m.

Facial Recognition: Frequently Asked Questions