At Least 12 Wrongful Arrests: Clearview AI's Casualty Count

TL;DR: At least twelve Americans have been wrongfully arrested after facial recognition software misidentified them. Angela Lipps, a Tennessee grandmother, spent six months in jail (the longest detention yet) for bank fraud committed 1,200 miles away in Fargo, North Dakota. She lost her home, her car, and her dog before anyone bothered to check her alibi. Previous victims include Robert Williams (30 hours in Detroit), Porcha Woodruff (arrested while seven months pregnant), and Randal Reid (six days in Louisiana for a theft in Georgia). Despite this track record, the US Army just extended Clearview AI’s contract through 2030. CBP paid $225,000 for access to 60 billion faces. ICE holds a $9.2 million contract. Nobody in Washington is stopping this.

UPDATE (March 2026): Angela Lipps, a 50-year-old Tennessee grandmother, has been added to this list. She spent 108 days in a Tennessee jail and another two months in North Dakota (nearly six months total) after Fargo police used facial recognition to identify her as a bank fraud suspect. She was 1,200 miles away when the crimes occurred. The ACLU calls this “the twelfth or so case we know” of wrongful facial recognition arrest in the US. Read the full story.

The Cases We Know About

These are the people who got lawyers, fought back, and made the news. The Legal Aid Society says the real number is “likely far higher” ^[1]. Facial recognition arrests rarely disclose the technology was used at all.

Trevis Williams, New York, 2025

In April 2025, NYPD arrested Trevis Williams outside his Brooklyn home for indecent exposure in Manhattan ^[2][3].

The actual suspect was 8 inches shorter. 70 pounds lighter. Didn’t look anything like Williams.

Cell phone data showed Williams was miles away when the crime occurred. None of that mattered until after he’d spent two days in jail.

The Legal Aid Society called it the seventh wrongful facial recognition arrest they’d documented in five years ^[2]. They also found that NYPD used other city agencies to run facial recognition searches that NYPD itself is barred from conducting. A workaround to dodge their own restrictions.

Williams was in the process of becoming a correctional officer at Rikers Island. After the arrest, they froze his hiring ^[3]. Even though the charges were dismissed. Even though it was the technology that failed.

Robert Williams, Detroit, 2020

The first publicly reported facial recognition wrongful arrest in America ^[4].

Robert Williams was arrested outside his Farmington Hills home in January 2020. His two young daughters watched. His neighbors watched. Police cuffed him in the driveway and took him to jail for allegedly stealing watches from a Shinola store in Detroit ^[4].

The “evidence”: Michigan State Police ran a blurry surveillance image through facial recognition. The system returned Williams as a match. A detective showed that match to a store security guard in a photo lineup. The guard picked Williams.

Williams spent 30 hours in an overcrowded cell. He was interrogated. He was fingerprinted. He was photographed. He was processed for a crime caught on camera: camera footage that plainly showed someone who wasn’t him ^[4].

In June 2024, Detroit settled the lawsuit. More importantly, the settlement forced the nation’s strongest police policies on facial recognition use ^[5]. Now Detroit requires independent evidence before any arrest based on a facial recognition match. They audit all facial recognition cases back to 2017. Officers get mandatory training on how the technology fails, especially on Black faces.

Porcha Woodruff, Detroit, 2023

Detroit police arrested Porcha Woodruff for a carjacking and robbery in February 2023 ^[6]. She was seven months pregnant.

The surveillance footage showed no pregnant woman. The eyewitness descriptions mentioned no pregnant woman. The facial recognition system didn’t care ^[6].

Woodruff was handcuffed in front of her two children, ages 12 and 7. She was held for 11 hours. She had contractions in the jail. Prosecutors dropped the charges after reviewing the evidence, or rather, the lack of it ^[6].

Detroit keeps making the same mistake. Same department. Same technology. Different Black victim.

Randal “Quran” Reid, Louisiana, 2022

The day after Thanksgiving 2022, Georgia state troopers pulled over Randal Reid and arrested him on a Louisiana warrant ^[7][8].

The charge: stealing thousands of dollars in luxury purses from a consignment shop in Metairie, Louisiana. Reid had never been to Louisiana.

Jefferson Parish detective Andrew Bartholomew ran the surveillance footage through Clearview AI. The system returned Reid as a match. Bartholomew wrote an arrest warrant citing “a credible source,” never mentioning that the source was facial recognition software ^[7].

No additional evidence. No alibi check. No verification that Reid was anywhere near Louisiana. Just a machine’s guess treated as probable cause.

Reid spent six days in jail before the warrant was rescinded. In 2025, Jefferson Parish settled the federal civil rights lawsuit for $200,000 ^[8].

Harvey Murphy Jr., Texas, 2022

Harvey Murphy Jr. was living in California when someone robbed a Sunglass Hut in Houston in January 2022 ^[9].

Months later, Murphy returned to Texas to renew his driver’s license. Police arrested him for the robbery. Macy’s facial recognition had identified him from surveillance footage ^[9].

He was 61 years old. He was 1,500 miles away when the crime happened. He sat in a Houston jail cell where three men sexually assaulted him ^[9].

The DA’s office eventually determined Murphy wasn’t involved. He filed a lawsuit against Sunglass Hut’s parent company EssilorLuxottica, claiming they relied on a facial recognition match from grainy footage to destroy his life ^[9].

Michael Oliver, Louisiana

Michael Oliver was accused of grabbing someone’s phone from their car. Police charged him with larceny based on a facial recognition match ^[10].

Oliver has dozens of tattoos. The actual suspect had no visible tattoos. The photos used to identify him showed a completely different person ^[10].

His case dragged on for months before a judge dismissed it. Only afterward did Oliver learn he’d been arrested because of facial recognition. Police never told him ^[10].

Angela Lipps, North Dakota/Tennessee, 2025-2026

The longest documented wrongful detention from facial recognition misidentification. Nearly six months ^[18].

On July 14, 2025, U.S. Marshals arrived at Angela Lipps’s Tennessee home with guns drawn. She was babysitting four grandchildren. Fargo, North Dakota police had issued a warrant for her arrest: bank fraud, tens of thousands of dollars stolen using a fake military ID ^[18].

Lipps had never been to North Dakota. The extradition flight was the first time she’d ever set foot in the state.

A Fargo detective ran surveillance footage through facial recognition. The algorithm returned Lipps as a match. He reviewed her social media and driver’s license photo, concluded she “appeared to be the suspect,” and filed charges ^[18]. He never called her. Never asked where she was during the robberies. Never checked her alibi.

Because she was classified as a fugitive from another state, Lipps couldn’t get bail. She sat in a Tennessee jail for 108 days. North Dakota didn’t retrieve her until late October. Police didn’t interview her until October 31, three and a half months after arrest ^[18].

Her bank records showed exactly where she was during the Fargo frauds: 1,200 miles away, making Social Security deposits, ordering Uber Eats, buying gas at Tennessee stations.

Charges were dismissed on Christmas Eve 2025. By then, she’d lost her home to unpaid rent. Lost her car to repossession. Lost her dog ^[18].

Lipps is white, the first documented white victim of facial recognition wrongful arrest in the US. The technology fails everyone.

The Pattern Nobody’s Fixing

Until Angela Lipps, every documented wrongful facial recognition arrest in the United States involved a Black person ^[1][4][6]. Lipps is white, the first documented white victim. But the technology’s racial bias remains stark.

National Institute of Standards and Technology testing found that facial recognition algorithms misidentify Black faces at rates 10 to 100 times higher than white faces ^[11]. The technology was trained predominantly on lighter-skinned faces. It reflects its training data.

Clearview AI disputes this. Their website claims “greater than 99% accuracy across all demographics” based on NIST testing ^[12]. But NIST also found that accuracy drops significantly on lower-quality images: exactly the grainy surveillance footage police actually use ^[11].

The police failure is consistent too. In every case:

Officers treated the algorithm’s output as near-certain identification
Basic investigation would have eliminated the suspect (alibis, physical descriptions, location data)
The technology was hidden from defendants: warrants cited “credible sources,” not facial recognition
The real perpetrator remained free while innocent people sat in jail

A Washington Post investigation found that police “probably could have eliminated most of the people as suspects before their arrest through basic police work” ^[13]. Check alibis. Compare tattoos. Follow DNA evidence at the scene. They didn’t.

Meanwhile, the Contracts Keep Growing

You’d think a track record of wrongful arrests would slow federal adoption. It hasn’t.

US Army 1st Special Forces Command: Renewed Clearview AI contract in February 2026, with options extending through March 2030 ^[14]
CBP: $225,000 contract for 60 billion face database access, going live September 2026 ^[15]
ICE: $9.2 million Clearview AI contract signed September 2025 ^[16]

The Army’s justification memo claims DOD “cannot rapidly analyze vast amounts of facial data” without Clearview ^[14]. CBP embedded Clearview into its National Targeting Center for “tactical targeting” ^[15]. ICE uses it for immigration enforcement.

All of this runs on the same database: 50+ billion images scraped from social media without consent. The same technology that put Robert Williams in a cell for 30 hours. The same system that couldn’t tell the difference between a 6’3” man and someone eight inches shorter.

What’s Not Being Done

No federal law regulates facial recognition use by law enforcement. None.

The ICE Out of Our Faces Act would ban ICE and CBP from using the technology. It won’t pass this Congress. Even if it did, it wouldn’t touch state and local police, who make the vast majority of facial recognition arrests.

Some states are moving. Virginia’s facial recognition restrictions take effect July 2026 ^[17]. Massachusetts has limited law enforcement use. A few cities have banned it entirely.

But most of the country operates with no rules at all. Police can run any face through Clearview, get a match, and call it “a credible source” in an arrest warrant. The defendant never knows. The judge never knows. The algorithm’s error rate is never disclosed.

Until it falls apart. Until someone like Robert Williams gets a lawyer. Until a settlement forces disclosure. By then, the damage is done.

What You Can Do

If you’ve posted photos online, you’re probably in Clearview’s database. You can’t opt out. The company ignores deletion requests outside of states with specific laws.

What you can do:

Know your rights: If arrested, ask whether facial recognition was used in the investigation. Police aren’t required to tell you, but defense attorneys can demand disclosure.
Document everything: Alibis, location data, timestamped photos. If you’re ever misidentified, this is your defense.
Support restrictions: State-level facial recognition bans are the only realistic path right now. Groups like the ACLU and EFF track legislation.
Limit exposure: The less your face appears online, the smaller your footprint in these databases. Not a perfect solution, but a real one.