Facial Recognition: Grandmother Jailed Months, Sues for $10M

Angela Lipps was in Tennessee when a string of bank thefts happened in North Dakota. Her own bank records prove it. She still spent months behind bars for those crimes, because a computer decided her face matched.

Angela's a fifty-year-old grandmother

Angela's a fifty-year-old grandmother. And if you've ever posted a photo of yourself online, her story touches your life too. That's where this one started. It started with a picture on her social media. Police in the Fargo area were hunting a woman who'd used a fake I.D. at several banks. Surveillance cameras caught her on video. According to ABC News, an A.I. facial recognition system compared that suspect to photos online, and it pointed at Angela. She was arrested in July of twenty twenty-five. She wasn't released until Christmas Eve. Now she's suing for ten million dollars. So one question runs through all of this. How does a computer's guess turn into handcuffs, with nobody checking along the way?

Start with what her lawsuit claims. According to the complaint, Fargo police knew facial recognition shouldn't be the only basis to charge someone. They also knew it shouldn't be the only reason to haul someone across state lines. The complaint alleges they did both anyway. It goes further. It alleges the department knew that officers using this tool without supervision could wrongly jail people. And it alleges the department chose not to supervise them. Those are allegations, and a court hasn't ruled on them. But the proof that cleared her wasn't hidden. It was sitting in her bank statements the whole time.

Months in jail. For an alibi that was one phone call away.

Most people picture What Is Identity Verification

Most people picture What Is Identity Verification Apple Cash Wants Id For 1 announcing a name. The software doesn't work that way. It hands back a list of possible candidates, ranked by how similar they look. Under police best practices, a trained human examiner then compares the faces and confirms or rejects each one. Some media reports suggest agencies are leaning harder on the software and less on those trained examiners. For a detective, that's the gap between a lead and a conclusion. For everyone else, it means a stranger's grainy photo could put your name at the top of someone's list.

So how often does the software get it wrong? That depends on who you are. According to the National Institute of Standards and Technology, known as N.I.S.T., false matches vary far more across groups than missed matches do. In its testing as of March twenty twenty-five, even the top system wasn't even across groups. It flagged West African women over sixty-five by mistake more than three hundred fifty times as often as middle-aged Eastern European men. Same software. Very different odds of being wrongly flagged.

Defenders of the technology point to a different N.I.S.T. finding. In certain tests, many algorithms get it wrong less than one percent of the time. Plenty land under half a percent. That's real. But those results come from clean, well-lit images, the kind taken at an airport passport check. Bank surveillance footage is nothing like that. It's often blurry. It's low resolution and washed out. The person's face is usually turned away from the camera. Half a percent sounds tiny. Unless you're the one in that half percent.