TTHE FALSE POSITIVE

Consider San Francisco, Flock's marquee success story. In March 2024, the city began building a 400-camera ALPR network. In 2025, Mayor Daniel Lurie called the system a "turning point for public safety." Earlier this year, Flock's CEO credited the cameras with a 25 percent drop in crime in 2025.

The city's own crime data tell a different story. Property crime in San Francisco began falling sharply in the second half of 2023 — months before the first Flock camera was installed. City officials acknowledged as much at the time. Violent crime followed the same trajectory. The 2024 and 2025 decreases that Flock's marketing takes credit for were, by the data, a continuation of a trend that predated the company's presence entirely.

The pattern is not limited to San Francisco, and it does not always run in Flock's favor — which is itself the point. In Springville, Alabama, which signed with Flock in 2019, the property crime rate was 419 incidents per 100,000 residents that year. By 2023 it had more than tripled, to 1,510 per 100,000. It has since eased somewhat but remains elevated above its pre-pandemic baseline. In Marietta, Georgia, which added cameras in 2017, the violent crime rate rose from 349 to 489 per 100,000 people over the following three years, before eventually falling back below its pre-pandemic level.

Flock claims its cameras accurately capture 93 out of every 100 license plates that pass by them. At the company's claimed scale of more than 20 billion individual reads per month, even that "high" accuracy rate implies well over one billion inaccurate reads every month — the overwhelming majority never relevant to any investigation, but a percentage inevitably are.

Hayward, California, June 2026. A Flock ALPR unit mounted at the intersection of B Street and Foothill Boulevard — one of thousands of such cameras now scanning traffic nationwide. Photo: RailTypes, CC0, via Wikimedia Commons.

An Institute for Justice review of media reports and court records identified at least 27 cases since 2018 — a running tally that has since grown to 33 — in which ALPR errors led to innocent drivers being pulled over, detained, or jailed. The majority occurred since 2023, tracking the pace of Flock's national expansion. In nearly two-thirds of these cases, officers did not realize their mistake until after they had already drawn and pointed their guns at innocent people.

About a third of the errors were pure machine failures — cameras confusing visually similar characters like 0 and O, or 2 and 7, or getting thrown off by dirt or a license plate cover. The rest were human error: officers entering the wrong data, forgetting to clear a resolved hotlist, or trusting an unconfirmed computer match without checking it against the actual photo. In Roseville, California, Flock sent local police 1,427 alerts over two years flagging cars as stolen or linked to a felony. Seventy-one percent of those alerts were misreads.

Behind the error-rate statistics are individual families whose worst five minutes are now part of the public record. In February 2026, a Flock camera in Sherwood, Arkansas misread a digit on an SUV's plate. Officers pulled the vehicle over and ordered the couple inside out at gunpoint — while their six-week-old baby sat alone in a car seat in the back. "Those cameras are placed everywhere, and they hit license plates," one of the officers explained afterward, uncuffing the couple once the error was found. "I'm not gonna say they're completely perfect, because, you know, that's modern technology."

A Flock Safety camera unit, mounted roadside. Networks like this now scan license plates in thousands of U.S. communities, feeding automated hotlist alerts directly to patrol officers.

The list of similar cases is long and geographically scattered. In Volusia County, Florida, a Flock capture mismatched an innocent driver's car to one linked to a fatal accident; he spent 13 days in jail before the error was caught. In San Diego, officers searching for a red Alfa Romeo tied to an attempted carjacking relied on Flock's "vehicle signature" matching technology — it flagged a different red Alfa Romeo five miles from the crime scene. All three occupants were arrested; one passenger spent nearly a month in jail over the holidays before officers realized their mistake. In Morristown, Tennessee, a camera misread an "O" as a "0" and officers detained two grandparents at gunpoint while their three-year-old granddaughter watched from the car. In Toledo, Ohio, a camera misread a "7" as a "2" — officers detained the driver at gunpoint, deployed a police dog against him, and held him for several hours.

Two cases in June 2026 involved the same underlying flaw compounding across state lines: automotive journalists driving loaner Range Rover test vehicles were tracked for days in Minnesota and Nebraska after a stolen-plate report entered into a national database was missing two digits that had been printed smaller than the rest of the plate. Flock's system matched on the truncated sequence in both cases; in Plymouth, Minnesota, the driver was boxed in by four squad cars outside a Kohl's before anyone checked the full plate against Flock's own photo.

Not every settlement stays quiet. In Aurora, Colorado, a misread minivan plate in 2020 led officers to detain a woman and her four children at gunpoint. The city later paid a reported $1.9 million settlement over the incident — one of the clearest price tags yet put on an ALPR error.

Flock's standard defense of its cameras is that they only record license plates — a narrow, largely anonymous data point. A false-arrest lawsuit filed in June 2026 against the town of Cumberland, Rhode Island undercuts that defense directly.

Pizzarelli v. Town of Cumberland disclosed that local police used a photograph taken by a Flock camera — not just a plate read — to identify a woman as a suspect in a vehicle pursuit. Officers used Facebook to cross-reference the image, then wrote in a police report that "a photo from the flock camera... matches [her] license photo." She was arrested. The identification was wrong.

The ACLU of Rhode Island sent a letter urging the Cumberland Town Council to cancel its Flock contract outright, and to disclose publicly when the expanded photo capability was activated, how often it has been used, and why it was never mentioned during the 2024 purchase approval. "These cameras are concerning enough when used just as license plate readers," the letter states, "but their danger is exponentially increased when expanded to take photos of people as well."

ACLU of Rhode Island executive director Steven Brown put it plainly: "This false arrest encapsulates the extremely serious privacy concerns that the ACLU and many others have been raising about Flock Safety and automated license plate reader cameras more broadly. This technology can do more than 'just record license plates,' and the secrecy surrounding its more invasive uses deserves special condemnation."

The Cumberland case did not happen in isolation. In the three weeks before the ACLU's letter, three other Rhode Island towns — South Kingstown, Narragansett, and Glocester — had already voted to cancel their own Flock contracts.

A fair investigation has to give Flock's strongest counter-evidence its due. In August 2026, the company published a working paper by two academic criminologists, Dr. Scott Mourtgos and Dr. Ian Adams, examining 216 law enforcement agencies as they adopted fixed ALPR systems at different times, using national FBI crime data. The finding: motor vehicle theft fell approximately 11 percent in the first year after camera deployment, with the effect concentrated in the jurisdictions hit hardest by car theft beforehand. Vehicle theft clearance-by-arrest rose 15.9 percent — a real gain in a category where, per the FBI, only 1 in 10 reported cases nationally end in an arrest.

Per Flock's own published FAQ, the company did not fund the study and says it had no role in shaping its design or results; its contribution was limited to providing deployment data and technical clarifications. If the design holds up under outside scrutiny, it would be the first study to estimate ALPR effects across this many agencies using a causal, staggered-adoption method rather than a single city's before-and-after comparison — a meaningfully more rigorous approach than the anecdotal percentages in Flock's own marketing.

It is also worth noting plainly what this study is not: it is a working paper, not yet published in a peer-reviewed journal, publicized on the company's own blog whose business depends on the result. That does not make its numbers false. It does mean readers should give it the same scrutiny they'd give any other unreplicated single study — including the studies IJ cites showing academic research on ALPR efficacy remains, in IJ's own words, "in its infancy."

The Institute for Justice has now cataloged more than 200 documented cases of ALPR abuse and misuse nationally — including more than 60 stalking incidents in which officers used the systems to track romantic interests, at least 33 wrongful detentions from camera or data errors, and more than 100 additional cases of officers disciplined or criminally charged for misusing the systems. IJ calls its own count likely "just the tip of the iceberg," noting there is no comprehensive national database of ALPR errors — the only jurisdiction that publishes regular error reports, Oak Park, Illinois, found that roughly a third or more of its Flock-triggered traffic stops ended in the driver being released due to data problems, with "no evidence whatsoever" the cameras had meaningfully aided any investigation. Oak Park canceled its Flock contract in August 2025.

Oak Park is not alone. Hundreds of communities nationwide have already canceled ALPR contracts, and Institute for Justice attorney Michael Soyfer is currently representing residents of San Jose and Norfolk in federal lawsuits challenging their cities' surveillance networks outright. In Rhode Island alone, four towns canceled or moved to cancel contracts within a matter of weeks in mid-2026.

None of this proves Flock's cameras never help solve a crime — the company's own commissioned study, on its own terms, suggests they measurably do, at least against vehicle theft. But "eliminate crime, full stop" was never a modest claim, and the record built since 2018 does not support it as written. What the record supports is a narrower, less marketable truth: a surveillance network that scaled to 20 billion monthly scans faster than any oversight system built to catch its mistakes, that took credit for a national crime decline already underway before it arrived, and that has, by its own accuracy numbers, generated well over a billion questionable reads a month — a handful of which have ended, at gunpoint, with a baby alone in a car seat, a three-year-old watching her grandparents get cuffed, or a month behind bars over the wrong red car.

The company's mission statement promises to eliminate crime. The record it has generated instead documents something else: how easily a genuinely useful tool becomes a liability the moment no one is required to check its work.

Every claim in this investigation is drawn from, and traceable to, the documents below. NFA's standing policy: if it is not in the record, it is not in the report.