He Sat in Jail 11 Months Because a Computer Thought His Face Looked Familiar

A man spent nearly a year of his life in jail for a murder he didn't commit — a murder that happened in 1998. The evidence against him? Mostly the fact that a computer said his face looked like someone in an old photo. Investigators had proof he wasn't the guy. They had fingerprint results from 2017 — seven years before they arrested him — that said, clearly, this is not your man. They arrested him anyway.

This is the story of Javier Lorenzano Nunez, and it is now a federal lawsuit. But honestly? It's also a warning to every person reading this at whatever hour you're reading it. Because the system that put him in a cell for 11 months is the same one that could, one bad day, point at your face.

Here's What Actually Happened

Phoenix investigators were trying to close a 1998 murder case — nearly 26 years cold. They ran an old photograph through a facial recognition database (think of it like a "find this face" search across millions of stored images) and got back 250 possible matches. Two hundred and fifty. They zeroed in on one: Lorenzano Nunez.

Not because other evidence pointed to him. Not because witnesses named him. According to ABC15 Arizona, investigators could not even find proof he had ever set foot in Arizona. But none of that stopped the arrest. He was extradited — meaning physically moved across state lines by law enforcement — and held for 11 months before the charges were quietly dropped.

Here's the part that should make your jaw drop. Arizona's own Department of Public Safety had already told investigators, in its own written guidance, exactly how this technology is supposed to be used. This article is part of a series — start with Blocked By A Bot Europe Just Gave You The Right To Demand An.

"Image comparisons are nonscientific and are intended for lead purposes only and should not be used as the sole basis for any decision." — Arizona Department of Public Safety, official guidance on facial recognition use

A lead. Not proof. Not grounds for arrest. A starting point that tells investigators where to look, not who to grab. Phoenix police apparently didn't get the memo — or got it and set it aside.

The Fingerprints Said No. Years Before.

This is the detail that keeps me up a little. Forensic fingerprint analysis — the kind of old-school, painstaking science that holds up in court — was completed in 2017. That analysis excluded Lorenzano Nunez from the crime scene. Excluded him. As in: these are not his prints. As in: wrong person.

Investigators had that result sitting in their files for seven years before they arrested him. According to ABC15's follow-up investigation, the detective handling the case also did not tell the grand jury that facial recognition was used, and didn't mention that witnesses had originally pointed to a completely different suspect years earlier. The grand jury — the group of citizens who decided there was enough evidence to charge Lorenzano Nunez — was working with a version of events that had some critical pieces missing.

His attorney put it plainly:

"They just assumed, based on two photos, that my client was the person they had been looking for, for 25 years." — Defense attorney for Javier Lorenzano Nunez, as reported by ABC15 Arizona

Two photos. A quarter century of an open case. And the pressure — human, institutional, entirely understandable — to finally close it.

This Isn't a Glitch. It's a Pattern.

You might think this is a one-weird-case situation. It isn't. Previously in this series: One Photo One Grudge One App The 10 Minute Nightmare Every P.

The ACLU has documented more than a dozen of these cases, and research from the Federation of American Scientists confirms what civil liberties groups have been saying for years: facial recognition produces higher rates of false matches for people of color, women, older people, and younger people. The technology isn't equally unreliable for everyone — it's specifically less reliable for the groups least likely to have power when something goes wrong.

Every time a new wrongful arrest surfaces, law enforcement agencies say the same thing: our policy is that facial recognition results are just a lead, not probable cause — which is the legal threshold (the minimum evidence needed) required to arrest someone. In theory, that's exactly right. In practice, research from the Justice Education Project shows officers treating algorithmic outputs as reliable conclusions — in some cases literally referring to an unverified match as a "100% match." That phrase isn't coming from the software. It's coming from a human who stopped questioning what the computer told them.

What "Verification" Is Supposed to Look Like

Look, nobody is arguing that face-matching technology should be thrown out entirely. Used correctly, it's a tool that helps investigators find leads faster — especially in cases where a witness description is the only starting point. The problem is the word correctly.

Correct use means: run the match, get a list of candidates, and then do the actual work. Talk to witnesses. Check physical evidence. Confirm someone was in the right place. Make sure your fingerprint results don't already exclude your suspect. (That last one feels almost too obvious to type, yet here we are.)

According to Hoodline's analysis of the Phoenix case, the warning signs were there at every stage. A different suspect had been identified by witnesses much earlier. Evidence that could have quickly ruled out Lorenzano Nunez was available for years before anyone knocked on his door. The face match didn't cause those failures on its own — it gave people with existing blind spots a reason to stop looking.

If you've ever wondered whether a photo — any photo — really tells you what you think it does, that instinct is exactly right. A face is not a fingerprint. It changes with age, lighting, angle, and camera quality. Matching a decades-old photo to a current database image is not the same as matching a fingerprint to a crime scene sample. The technology itself is built on probabilities, not certainties, and the people running it don't always communicate that distinction clearly. One thing worth knowing: before any image-based identification technology is used to make a real decision about a real person, the baseline question should always be — what else confirms this? What does the non-visual evidence say? Up next: Liveness Detection Selfie Id Verification Explained.

Lorenzano Nunez wasn't freed because someone finally did brilliant detective work. He was freed because the forensic evidence that had already cleared him — sitting in a file for seven years — eventually made it impossible to keep pretending the case was solid. That's not justice. That's a system correcting itself after doing serious, irreversible damage to a real person's life.

He is now suing Phoenix PD and the Maricopa County Attorney's Office. Whatever happens in court, the lawsuit won't give him those 11 months back.

Here's the question that actually keeps this story alive beyond the headlines: the grand jury that approved his arrest didn't know facial recognition was involved. They couldn't ask whether the match was strong or weak, whether 249 other people were also flagged, or whether fingerprint evidence had already cleared the defendant. They decided based on what they were told. Which means the single most important safeguard in this whole system — a room full of ordinary citizens deciding if there's enough evidence to charge someone — was working with incomplete information.

If that doesn't make you want to know exactly what evidence is being used to make decisions about people who look like you, or your kid, or your parent — I'm not sure what will.