Police departments are embracing AI-powered video surveillance. This company is leading the effort

A red Hyundai SUV makes its way through downtown San Francisco on a bright August afternoon. The car looks innocuous, but a day earlier, police had linked it to the break-in of another car and added it to a so-called hot list of wanted vehicles. Now the city’s license plate recognition cameras, or LPRs, are stationed at major intersections, ready and waiting.

One of those LPRs photographs the Hyundai’s plates, matching it with the car on the hot list and sending an automated alert to the San Francisco Police Department’s command center. Officers launch a drone to monitor the car. From afar, they watch as two people in hoodies and gloves emerge from the Hyundai, smash the window of a parked car, and wrestle suitcases from the back before driving off. Minutes later, police spike the Hyundai’s tires, arrest the suspects, and confiscate their handguns.

Welcome to the era of tech-enhanced policing, the stuff of public safety advocates’ dreams—and civil liberty advocates’ nightmares. It’s a world where cameras patrol the streets and software helps solve crimes, and it’s being hastened by a small number of surveillance companies, including the one behind the LPR that snagged the Hyundai: Flock Safety. The Atlanta-based startup, which also sells security cameras, first-responder drones, and a full suite of command-center software tools, signed a deal with San Francisco in March 2024 to install and operate 400 LPRs. By the time of the Hyundai incident in August, the city had already installed more than 200 of Flock’s solar-powered, internet-connected, AI-enhanced infrared cameras, placing them at intersections across its 46.9 square miles.

Signing with Flock Safety may seem a surprising move for such a legendarily progressive city. In just the past decade, San Francisco has decriminalized drug possession, eliminated cash bail, and reduced its jail and prison population. But as the situation on the city’s streets deteriorated during and after the pandemic, leading to open-air drug markets, sprawling homeless encampments, and flagrant burglaries, residents and small-business owners started to question some of those policies.

In 2022, residents voted to recall Attorney General Chesa Boudin, who had become a potent symbol of criminal justice reform (and its failures). Intent on avoiding a similar fate, Mayor London Breed threw her weight behind a ballot measure known as Proposition E that grants police greater access to surveillance technology. Voters passed it in early March and, two weeks later, Breed announced a $3.9 million deal with Flock. Breed lost her reelection campaign anyway, defeated in November by Levi Strauss heir Daniel Lurie, who voiced a similar stance on crime.

“Technology to help address crime in the city has been a game changer,” Breed told me over the phone in October, when the mayoral race was in a dead heat. “This is helping us step up the work that we need to do to ensure not just safety but that folks who would try and commit crimes in our city know that they’re not going to get away with it.” That perspective is shared by many of the city’s leading technocrats, including Garry Tan, president and CEO of San Francisco-based Y Combinator, who led Flock Safety’s seed round in 2017 while running venture firm Initialized Capital. He is one of several prominent business leaders pushing for an increase in what they describe as “common sense” policing and public safety measures.

But despite the narrative of San Francisco as a crime-ridden city, the data is more nuanced. In 2023, according to the SFPD, violent and property crime both fell below pre-pandemic levels. This trend extends to many other U.S. cities, the majority of which have seen crime rates drop to roughly 2019 levels—which are themselves lower than rates in the early 1990s, according to the Council on Criminal Justice.

But there is one prominent exception: motor vehicle thefts, which have more than doubled across the country over the past five years. Many go unsolved—a problem so severe that in some cities, volunteer detectives patrol the streets. (A private Facebook group where volunteers in Portland, Oregon, coordinate their efforts has nearly 22,000 members.) Amid this backdrop, short-staffed police departments increasingly view LPRs like Flock’s as an essential tool.

Flock Safety has crafted its sales pitch to meet the moment, presenting LPRs as a way to combat vehicle theft and expanding into other aspects of policing from there. Founded in 2017, the company has raised more than $380 million in venture dollars and was most recently valued at $4 billion. Flock’s first product was an LPR that paired mounted cameras with AI-powered image recognition software; it originally sold the devices for a 10th of what its competitors charged.

Over time, the company has added a full suite of companion products to its “flock,” including security cameras and gunshot-detection devices. In October, it acquired drone developer Aerodome, which specializes in first-responder drones. After attracting local customers in and around Atlanta, Flock now serves cities from coast to coast, including Oakland, California; Austin, Texas; and Providence, Rhode Island. Homeowners’ associations, schools, and malls are customers, too.

Incumbents in the video security and surveillance space—a market worth around $60 billion—tend to emphasize safety. “Solving for safer” is how Motorola Solutions, which makes video security systems, describes its mission. “Protect more lives in more places,” proclaims Axon, maker of Tasers and police body cameras. Flock goes a step further, promising to eliminate crime altogether. “I actually don’t know why crime has to exist,” says Garrett Langley, cofounder and CEO of Flock Safety, citing low crime rates in countries like Japan as inspiration.

Langley believes that successfully solving and prosecuting more crimes will eventually dissuade anyone contemplating a “crime of opportunity” from acting in the first place. Last February, his company published a study purporting to show that 10% of reported crime in the U.S. is currently being solved with Flock Safety. The study was only preliminary, but Flock continues to reference the finding in its marketing. The 10% claim performs a neat trick: Tying Flock’s technology to quantifiable outcomes makes the company’s products feel less like mass surveillance and more like targeted intervention.

Langley is a Georgia native, but after participating in Y Combinator and raising capital from the likes of Andreessen Horowitz and Tiger Global Management, he’s at ease with Silicon Valley’s fondness for head-spinning ambition. “Here’s the world that I hope we get to,” he says. “In some near future, we can walk up to a mayor and city council and say, ‘Would you like the No Crime package?’ And then, ‘Here’s the set of products that can guarantee that crime will go to zero in the next five years.’ Maybe, for various reasons, you’re okay with nonviolent crime existing. ‘Great, here’s a package for that.’ I’m so convinced that crime is avoidable.”

Criminologists blanch at Langley’s simplification of how policing works and what motivates criminals. Civil libertarians, meanwhile, object to his evangelizing of surveillance-based policing, which they say violates the Fourth Amendment’s prohibition against “unreasonable searches.” Indeed, the American Civil Liberties Union has recommended that local communities reject contracts with surveillance companies that decline to subject their products to independent evaluation, and it has raised specific concerns about Flock Safety. Flock’s claim to be “best in class” is “puzzling and laughable,” the ACLU wrote in an article on the organization’s website in March, if it won’t allow independent testing of its LPRs’ accuracy.

But that hasn’t stopped some 5,000 police chiefs, homeowners’ associations, superintendents, and others from buying Flock’s systems. Individually, each of them wants some sort of assurance that their constituents, and their jobs, are safe. Collectively, through the billion vehicles per month that their shared security and policing systems photograph, they are establishing a new normal for public surveillance.

Gray skies threaten rain as Langley and I pull up to police headquarters in Dunwoody, Georgia, an Atlanta suburb with roughly 50,000 residents and a string of corporate office parks. We head inside the shared municipal building, with its cement cornices and columns, and take seats in the darkened, windowless room that serves as the police department’s command center. One long wall is blanketed by 12 screens, which together display maps showing where Dunwoody’s Flock devices are located, alongside real-time alerts from its LPRs (“stolen vehicle,” “shoplifting suspect”). Each time a car passes an LPR, Flock’s software parses the plate, translating the image into text, then looks for plate matches in the National Crime Information Center, a database managed by the FBI, as well as in any locally generated hot lists.

Patrick Krieg, who leads Dunwoody PD’s administrative services and investigations, looks the part of a no-nonsense cop with a trim brown beard, black polo shirt, and a can of Red Bull in hand. Krieg says when Flock Safety initially pitched the department in 2019, he thought its tech would be a better fit for homeowners’ associations and small businesses, which don’t necessarily need to respond in real time or weave together a wide array of tools.

“We thought this was a low-end product that really wouldn’t service a law enforcement application,” he says. But he agreed to a trial: He’d place one Flock LPR on a speed trailer, the portable device that police use to display the speed of passing cars. “That one camera in two days solved a commercial burglary group running up and down the East Coast on a federal case,” he says.

From that initial experiment, Dunwoody added 20 LPRs to identify vehicles entering the city, then another 20 to capture vehicles exiting. Then it installed 25 more at central intersections, including near Dunwoody’s bustling Perimeter Mall, home to Nordstrom and Macy’s. To date, Flock’s LPRs have helped with cases involving shoplifting, stolen vehicles, human trafficking, and more.

“Big brother isn’t really a term we run into as much anymore—and you think we would, because look at this map,” Krieg says, gesturing to the screens around him. Colored icons representing Flock’s LPRs, cameras, and gunshot-detection devices blanket maps of the city. “But I think [local people] have quickly realized we don’t have the time, resources, or, honestly, interest in watching” video feeds. Instead, his officers have outsourced that vigilance to Flock’s real-time software, freeing themselves up to react when the system flags something. “What we want to do is respond to incidents accordingly and strategically.” (Flock lets each customer decide how long to keep its LPR data; Dunwoody uses the default setting of 30 days.)

Krieg points out that Perimeter Mall has 1,200 cameras of its own, a handful of which feed into Dunwoody PD’s Flock system. And yet when there was a shooting at the mall a few years ago, he says, “Flock was the single source of evidence we pulled from.” The shooter was arrested in less than an hour. The city’s crime rates have fluctuated over time, but one number has soared in recent years: arrests of wanted persons. In 2019, Dunwoody PD made 166 such arrests. Last year, it made 337.

Langley, an avid runner who has the all-angles build of a teenager, leans back in his chair as he listens. Flock’s Series A pitch deck, which the company shopped around to investors in early 2018, didn’t even mention law enforcement as a potential customer. When a salesperson got an inbound request from a Texas police chief back in 2018, Langley hesitated. Only the size of the proposed contract—for 50 LPRs—swayed him.

Flock’s investors were “furious,” he says. Trae Stephens, cofounder of Anduril and a partner at Founders Fund, an early backer of Flock Safety, told Langley to steer clear of local government customers because of their modest budgets and complex needs. Why not focus on homeowners’ associations and private businesses, as Flock originally intended? “‘Trae, I hear you, but they’re calling and giving us money,’” Langley recalls saying. Today, police departments are the company’s primary customer.

The shadow on the horizon is Flock’s increasingly sprawling network of LPR data, which can be shared among customers, as well as with federal authorities. One lone LPR may not threaten privacy, but thousands of interconnected LPRs, placed in thousands of communities, start to raise alarms. Today, the concerns—and lawsuits—of civil liberty groups are growing just as quickly as Flock’s presence across the country.

In some respects, Langley shares their concerns. He has grand plans for his company, but a circumspect view of technology in general. “There was a moment in time, let’s call it early 2000s, when we talked about technology as if there would be no negative consequences,” he says. “That’s a pretty idealistic state [of mind] that’s largely now been dismissed.” Flock Safety, he insists, was founded with “wide-open eyes.” The company provides customers with self-auditing tools for compliance with local regulations; self-audits identified two cases in which officers conducted inappropriate searches using Flock’s data. More broadly, Langley wants objective and actionable to be the words people use to describe Flock products.

“Before LPRs, really before Flock, patrol officers just pulled over suspicious vehicles,” he says. “What are suspicious vehicles? They’re beat-up old cars.” He mentions that the Fort Worth, Texas, police department, since adopting Flock LPRs, has seen a decrease in the number of vehicle stops but an increase in the number of related arrests. He says it’s because they’re stopping “exact matches” rather than subjectively suspicious leads.

Testimony like that does little to quell the objections from civil liberty advocates. “It can be true that automated systems have less bias, but often that is true only in the narrowest sense,” says Jay Stanley, a senior policy analyst with the ACLU’s Speech, Privacy, and Technology Project. “In fact, the technology can make the consequences of bias worse.”

Stanley points to Ferguson, Missouri, where police killed teenager Michael Brown in 2014. Before the shooting, “Ferguson made up a chunk of their budget by nickel-and-diming disproportionately poor Black people on minor traffic violations, leading to an escalating series of fines and missed court dates that would suck people into the criminal justice system,” he says, citing a Justice Department investigation of the city’s practices. “A license plate recognition system is the perfect technology for implementing that kind of nightmarish policing regime.”

At the same time, seemingly mundane LPR mistakes can lead to potentially deadly encounters with police. In one prominent example, Brian Hofer, the chair of Oakland, California’s Privacy Advisory Commission, was stopped by police in 2019 when an LPR flagged a car he had rented for a Thanksgiving trip. The car had been stolen but was later recovered; no one had removed it from stolen vehicle databases. (Hofer sued, receiving a $30,000 settlement.)

In Piedmont, California, last year, police detained a man at gunpoint after a thief swapped their cars’ plates and an LPR flagged the stolen one. Also last year, an Ohio Lyft driver sued Flock Safety after he was detained by police for 40 minutes; because of a hot list of cars distributed by Flock, police believed he was a human-trafficking suspect. (The case was dismissed.)

Other scenarios raise significant privacy questions. U.S. Immigration and Customs Enforcement, or ICE, at one point had a contract with another LPR vendor that gave the agency access to regional license plate data. Shared data could be used to identify a woman crossing state lines to obtain an abortion, or a cannabis entrepreneur delivering orders to customers. Anywhere there is a significant gap between state laws, LPRs could be weaponized.

Plus, as LPRs get cheaper and more ubiquitous, the network they form starts to look less like an alert system and more like one built for mass surveillance. What’s the practical difference, critics ask, between following a car via a GPS tracker and following it via LPRs on every corner? In October, the Institute for Justice, a civil liberties organization, sued Norfolk, Virginia, for what it described as “warrantless surveillance,” a violation of the Fourth Amendment, citing the city’s Flock system, comprised of “172 unblinking eyes.”

As for Flock, Langley says he’s agnostic about how the company’s products are used, as long as they’re deployed legally. “The beauty of the products we’ve built is we get out of the way,” he says.

Even when LPRs function as designed, criminologists question their effectiveness. “There’s no strong evidence that these are really effective crime-fighting tools or investigation tools,” says Bryce Peterson, an adjunct lecturer at John Jay College of Criminal Justice and a research scientist at the Center for Naval Analyses, a safety- and security-focused think tank. He’s heard cops anecdotally point to examples of cases they’ve solved with help from LPRs. But in policing, controlled experiments are difficult to conduct.

In a recent evaluation of surveillance systems in Milwaukee, which has adopted fixed cameras and LPRs, Peterson found some evidence of improved clearance rates, which indicate that an investigation has been resolved. But crime rates went up, too, perhaps because the system was detecting more crimes. “It’s hard to isolate the effect of the LPR,” he says.

For Flock customers, most of whom are focused on their local communities, those academic findings may be less important than the perception that law enforcement has reclaimed control of the streets. “I don’t know what the data says one way or another,” San Francisco Mayor Breed told me when asked about research on LPRs’ mixed track record, “but I do know overall crime is down in San Francisco.”