Disturbing History - Privacy Not Included

Episode Date: August 16, 2026

You clicked “I Agree.” You never clicked anything for the camera on the pole outside your kid’s school. On May 9, 2025, a deputy with the Johnson County Sheriff’s Office in Texas typed “had an abortion, search for female” into a license plate reader network. That single search swept across 83,000 cameras connected to 6,809 law enforcement networks, including cameras in states where the conduct being investigated was legal. No warrant. No judge.Just a login and a sentence in a search box.In this episode, I dig into two surveillance systems most Americans barely realize they’re part of: the computer you’re driving and the network watching it move.We start inside the car, tracing vehicle data collection from the earliest event data recorders to the 2006 federal rule that standardized them. We revisit the Massachusetts lieutenant governor whose dashboard data contradicted his account of a 108-mph crash, then move into modern infotainment systems capable of copying contacts, text messages, call logs, and location history—and sometimes keeping that information long after the car changes hands.We also examine the Berla iVe forensic systems purchased by U.S. Customs and Border Protection for $456,073, Mozilla’s 2023 privacy review that failed all 25 car brands it examined, and the March 2024 revelation that General Motors had been sharing driving data with LexisNexis and Verisk, in some cases recording behavior at intervals of just a few seconds.Then we move outside the vehicle.Automatic license plate recognition began in 1976 with the British Home Office, became part of London’s Ring of Steel in 1993, and decades later evolved into a solar-powered camera network marketed to police departments, neighborhoods, businesses, and homeowners associations. That company is now valued at $8.4 billion, with more than 120,000 cameras operating across more than 5,000 communities.And the technology can work remarkably well. Plate-reader networks helped locate the Brown University shooting suspect and, in Galt, California, helped police find a kidnapping victim within five minutes.But the same system has also documented a very different kind of use. According to the Institute for Justice’s database of 147 incidents, a Georgia deputy searched one woman’s plate 1,639 times. A Florida officer searched for his estranged wife’s vehicle more than 700 times. In Milwaukee, a detective assigned to investigate another officer’s misuse of the system was later discovered to be misusing it himself. Accuracy creates another problem. Roseville, California police found that 71 percent of their plate-reader alerts came from misreads. In Morristown, Tennessee, grandparents were held at gunpoint after a camera mistook the letter O for the number zero.And then there’s Norfolk, Virginia.Discovery in a federal lawsuit revealed that among roughly 230,000 searches, the three most common justifications officers entered were simply “cj,” “stolen,” and “inv.”One officer repeatedly typed “la la la la” as the stated reason for searches—and apparently nobody noticed.Sixteen years in law enforcement taught me what a plate hit can be worth on a bad night. It also taught me what can happen when a powerful investigative tool operates with little oversight. Now the legal ground underneath these systems may be shifting. The Supreme Court’s June 29, 2026 decision in Chatrie v. United States undercut the reasoning Norfolk relied on to prevail earlier in its case, and that fight is now before the Fourth Circuit.Email BrianJoin Our FREE NewsletterGet Brian's Books Leave Us A VoicemailVisit Our WebsiteHave a forgotten historical mystery, disturbing event, unsolved crime, or hidden conspiracy you think deserves investigation?Send your suggestions to [e-mail] History is a dark history podcast exploring unsolved mysteries, secret societies, historical conspiracies, lost civilizations, and the shadowy stories buried beneath the surface of the past.Follow the show and enable automatic downloads so you never miss a deep dive into history’s most unsettling secrets.Because sometimes the truth is darker than fiction.

Some stories were never meant to be told. Others were buried on purpose. This podcast digs them all up. Disturbing history peels back the layers of the past to uncover the strange, the sinister, and the stories that were never supposed to survive. From shadowy presidential secrets to government experiments that sound more like fiction than fact, this is history they hoped you'd forget. I'm Brian, investigator, author, and your guide through the dark corner.

of our collective memory. Each week I'll narrate some of the most chilling and little-known tales from history that will make you question everything you thought you knew. And here's the twist. Sometimes the history is disturbing to us. And sometimes, we have to disturb history itself,

just to get to the truth. If you like your facts with the side of fear, if you're not afraid to pull at threads, others leave alone. You're in the right place. History isn't just written by the victors. victors. Sometimes it's rewritten by the disturbed. On the 9th of May, 2025, a deputy with the Johnson County Sheriff's Office in Texas sat down at a computer and ran a search across a national

network of license plate cameras. The system asked him to type a reason. He typed, had an abortion, search for female. That single query reached across more than 83,000 cameras belonging to nearly 7,000 separate law enforcement networks, including cameras in Washington and Illinois, states where what she'd done was legal. He didn't need a warrant. He didn't need a judge. He needed a login and a sentence in a text box. I want to start by telling you what this episode is not. Over the years, I've done shows about the phone in your pocket, about the apps you install and the permissions you hand over without reading a word, about terms and conditions that run longer than a novel, written by lawyers who are paid to make sure you don't finish them, and about the

fact that every single one of us has sat there with a thumb hovering over a blue button that says, I agree, and pressed it, because the alternative was not using the thing we just paid for. I've made my piece with a certain amount of that. You click, you consent, you get something in return. That's a bargain, even if it's a lopsided one. You can at least point to the moment you made it. This is different, and that's why I wanted to spend a few weeks. digging into it because there's no button you press to agree that a camera bolted to a utility pole three blocks from your house would photograph your car every time you drove past it log the time log the direction and store it in a

database that a deputy in another state can search there's no terms of service for the intersection outside your kids school and most people have no idea that the vehicle sitting in their driveway right now is keeping a record of where they went how fast they got there how hard they hit the brakes when they arrived and which phone was paired to the stereo when it happened. The car isn't a car anymore. It's a computer that happens to have wheels. And computers keep logs.

That's what they do. That's all they do. I spent 16 years in law enforcement. A good chunk of that with the Atlanta Police Department. And I want to be honest with you up front about my bias here. Because I think you deserve to know it before I start telling you what I found. I have stood in front of a whiteboard at three in the morning with nothing. No witnesses, no camera footage, no plate, nothing but a body and a family waiting on a phone call.

If somebody had handed me a tool that could tell me which cars passed that corner in the 20 minutes before it happened, I'd have taken it and said thank you. So I understand the appeal. I understand it in my bones. What I did not expect going into this was how much of the record would consist of that same tool being pointed at people who hadn't done anything at all. Let's start with the machine itself, because the surveillance built into your car got there first, and almost nobody noticed it happening.

The story usually starts with General Motors in the mid-1970s, with a device meant to help engineers understand whether their new airbags were firing correctly. It recorded a few fractions of a second around a crash. It was a diagnostic tool, not an evidence tool, and for a long time, that's all it was. By the 2000s, the technology had quite, quietly spread through the entire industry, and the federal government stepped in to standardize it rather than to require it. In August of 2006, the National Highway Traffic Safety Administration issued a rule under Part 563 that told manufacturers what their event data recorders had to capture if they installed them at all. Speed, engine throttle, brake application, steering input, seatbelt status, change in velocity during impact, airbag deployed,

deployment timing. Up to around 30 separate data elements recorded in a loop that overrides itself until something violent enough happens to freeze the file. The rule didn't force anybody to install one. It didn't have to. By the time the compliance date arrived, the agency estimated that roughly 96% of new passenger vehicles already had the hardware. The market got there without a mandate because the data was useful to the people building the cars. Beginning with the 2011 model year, manufacturers had to disclose the recorder's existence in the owner's manual, which is a wonderful piece of legal theater when you think about it. The disclosure exists. It's in the glove compartment. It's on page 400 and something. In the back, past the tire pressure chart.

The first time most Americans learned any of this, it was because a politician got caught. In February of 2004, Governor John Baldacci of Maine went off an icy highway and a state-owned Chevrolet Suburban. Investigators pulled the recorder. The existence of that data and the questions it raised about who owns it pushed the main legislature to act. And by August of 2006, the state had a law on the books saying nobody could download that information without the owner's permission or a court order. Then in 2011, the lieutenant governor of Massachusetts, Timothy Murray, wrecked a government car. He said he'd lost control on an icy road. He said he'd been wearing his seat belt. He said he'd been driving the speed limit. The event data recorder in that

vehicle said he was traveling at 108 miles an hour and was not belted in. That's the part that should have made everybody sit up. Not the crash. The fact that a man told a story and a machine in the dashboard contradicted him and the machine won. Courts have been treating that data as evidence for a long time now. In Virginia alone, the appellate record includes a case where a driver was convicted of involuntary manslaughter after his vehicle's recorder showed him traveling between 121 and 122 miles an hour just before impact another where the data established the defendant wasn't wearing a seat belt and never attempted to break or steer away another where a state trooper testified that the brakes weren't fully engaged and that testimony helped uphold a drunk driving conviction i don't have a problem with any of that

A man drives 120 miles an hour into another human being, and the car says so, and a jury hears it. That's the system working. But here's what I want you to hold on to as we go, because it matters for everything that follows. The event data recorder is tiny. It captures a handful of seconds, five seconds, 30 seconds, depending on the manufacturer and the event. It's a snapshot of a catastrophe, and it stays inside the vehicle unless somebody physically connects to. it. That's the old technology. That's the polite version. What's in your car now is something else entirely. The modern vehicle has somewhere in the neighborhood of a hundred separate

computers in it. A decade or so before that, the number was closer to 10. Those computers talk to each other constantly. They talk to your phone, and increasingly, they talk to a cell tower. One frequently cited estimate in the forensics industry puts the output of an American car at roughly 25 gigabytes of data per hour of operation. We're talking about a machine generating more information about you in a single commute than most people generated about themselves in a decade before smartphones existed. There are two systems doing the heavy lifting. The first is telematics. That's the piece that connects your vehicle to a network. General Motors calls its version OnStar. Ford calls its version sync. Every major manufacturer has one.

Telematics is why your car can call an ambulance if you're unconscious, why the dealer knows your oil life, why your phone can unlock your doors from an airport in another state, and why the manufacturer knows where you parked last Tuesday. The second is the infotainment system. That's the screen in the dash. And the infotainment system is the one that should worry you, because it doesn't just know about the car.

It knows about you. When you pair your phone to the stereo so you can take calls hands-free, The vehicle copies things over. Contact lists. Call logs. Text messages. Sometimes photos. Sometimes email.

It caches navigation destinations and the routes you actually took, not just the ones you asked for. It timestamps door openings and closings, gear changes, ignition cycles, and whether the vehicle was in park when a particular device connected. And it keeps that. It keeps it after you unpaer. It keeps it after you see.

sell the car. There's a company in Maryland called Burla Corporation that built the tool law enforcement uses to get it out. The product is called IVE. It launched in 2013 and the Department of Homeland Security's Science and Technology Directorate supported its development starting that same year, then helped connect the company to state and local police departments. When IVE launched, it supported around 80 vehicle models. By 2017, the company was claiming support for more than 6,700 globally. It pulls navigation data, location histories, connected device data, call histories, text messages, contacts, photographs, video, social media artifacts, and vehicle events down to the individual door opening. The Intercept obtained a contract in 2021, shared by the advocacy

organization Magent, showing that customs and border protection paid a Swedish extraction firm $456,073 for a bundle that included five IV vehicle forensics kits. An internal document said the kit would be critical, because it could provide evidence not only about the vehicle's use, but information obtained through mobile devices paired with the infotainment system. The document also noted that IVE was the only tool on the market that could do it. The following year, Forbes reported that customs and border protection spent more than $380,000 on the software in a single August purchase, roughly eight times its previous largest single buy. Immigration and customs enforcement, which had been buying

Berla tools and training since 2010, spent $500,000 that September, more than double its previous record. The San Diego Police Department's own surveillance disclosure page, describes what its Burla software does in plain language. It takes data stored in the infotainment and telematics systems, including vehicle events, location data, and data from connected devices. I keep coming back to that last phrase. Data from connected devices. Your phone is protected in ways the courts have spent years working out.

Your car's copy of your phone is a different question entirely, and it lives in a box under the dashboard that most people have never thought about once. Now, the obvious response to all of this is that it catches bad people, and it does. Prosecutors have used vehicle data to put suspects at crime scenes, to break alibis, and to reconstruct the last movements of missing persons. In one federal case, prosecutors relied on testimony from a Burla employee who told the jury that the Federal Bureau of Investigation had asked him to extract data from the defendant's vehicle. An agency digital forensic examiner followed him on the stand and testified that his analysis of the car's hard drive placed the vehicle

at the scene at 909 in the evening. Witnesses had testified to hearing gunshots between 915 and 930. That's tight. That's the kind of timeline a detective dreams about. The jury deadlocked. I include that because it's the honest ending and because it's the part that gets left out when this technology gets sold to a city council. Data isn't the same thing. as proof. A car being somewhere, isn't a person being somewhere. A door opening at a particular second doesn't tell you who opened it. Any investigator who's worked a real case knows that the record and the truth are cousins, not twins, and the gap between them is where the work lives. The problem is that the gap keeps getting narrower in the courtroom and wider in the public imagination,

and the people selling these systems are not in a hurry to correct that. Instead, In September of 2003, the Mozilla Foundation published a report that I'd recommend to anybody who owns a vehicle built in the last 10 years. They run a consumer guide called Privacy Not Included, and they've been reviewing connected products for seven years. That year, they reviewed 25 major car brands. All 25 failed. Every single one.

It was the first time in the guide's history that an entire product category flunked without exception, and Mozilla wrote that cars were the worst category they'd ever reviewed for privacy. The specifics are where it gets strange. Nissan ranked at the very bottom. Its privacy policy stated that the company could collect information including sexual activity, health diagnosis data, and genetic data. It did not explain how. The policy said the company could share and sell consumers' preferences, characteristics, psychological trends, predispositions, behavior, attitudes, intelligence abilities, and aptitudes to data brokers, law enforcement, and other third parties.

Kia's policy said the company could collect information about a customer's sex life. Six of the 25 brands listed genetic information or genetic characteristics as data they could collect. Volkswagen collected demographic information like age and gender, along the with driving behaviors including seatbelt and braking habits and used it for targeted marketing. Toyota's policies were spread across 12 separate documents. Mercedes-Benz shipped certain models with TikTok pre-installed. Subaru's policy took the position that passengers in a vehicle using connected services had consented to the collection and possible sale of their personal information,

which is a remarkable thing to say about somebody who got a ride to the airport. Hyundai's policy said it could share data with government and law enforcement based on formal or informal requests. Mozilla concluded that 84% of the brands they reviewed, share or sell customer data collected through the vehicle and its associated apps. Now, in fairness, and I do want to be fair here, the manufacturers pushed back. Nissan North America said those categories appear in the policy to comply with a growing patchwork of state privacy laws, some of which require companies to notify customers about information they might collect inadvertently. Nissan specifically denied intentionally collecting data about sexual activity.

Kia said it does not and has never collected sex life or sexual orientation information from vehicles or consumers in the context of providing its connected services, and that many of the listed categories were included for compliance with the California Consumer Privacy Act. I find that explanation partly persuasive and partly maddening. Partly persuasive because corporate privacy policies really are written defensively, listing everything a company might conceivably touch so that nobody can later claim they weren't warned. Maddening because that's the entire problem.

The policy is the contract. If the contract says the company may collect and sell information about your sex life, then the company has reserved the right to collect and sell information about your sex life and a press statement saying they don't currently do that is not the same as a promise that they never will. You can't hand somebody a legal document that grants you a power and then act insulted when they read it out loud.

Which brings us to the part of this story where a company got caught, and the reason we know about it is one reporter deciding to actually read her own paperwork. In March of 2024, Kashmir Hill at the New York Times published an investigation into General Motors. What she found was that the company had been collecting detailed driving data from millions of Chevrolet, Buick, GMC, and Cadillac vehicles through a feature called OnStar Smart Driver and sharing it with two data analytics firms, LexisNexis Risk Solutions and Verisc Analytics, which in turn sold driver risk profiles to insurance companies.

The data wasn't a summary. It was granular. Every trip. Every instance of hard braking. Every rapid acceleration. Every time the vehicle was driven late at night, every time it exceeded a speed threshold. Drivers were finding out about it when their insurance premiums went up and they couldn't figure out why. One of them was a Florida man named Romeo Chico. He bought a Cadillac XT6 in 2021, downloaded the manufacturer's app and according to his complaint, specifically declined to enroll in smart driver. He alleged the data was collected and shared anyway.

He filed suit on the 13th of March, 2004. Seven days later, on the 20th of March, General Motors announced it had terminated its relationships with both LexisNexis and Verisk. A company spokesman told the Detroit Free Press that as of that date, smart driver customer data was no longer being shared. That customer trust was a priority, and that the company was actively evaluating its privacy processes. He declined to say how long the arrangement had been in place. or why it was ending at that particular moment, which, I'll say plainly, was the week after it appeared in the New York Times. In August of 2024, the state of Texas sued General Motors and OnStar, alleging a years-long scheme to collect and sell driver data without knowledge

or consent, using telematics installed in most model year 2015 and later vehicles, and alleging the company took in millions in lump sum payments for it. In January of 2013, 2025, the Federal Trade Commission announced a consent order. In the statement accompanying it, then Chair Lena Kahn said the company monitored and sold people's precise geolocation data and driver behavior information, sometimes as often as every three seconds. That's not a record of your trips. That's a record of your life. Sampled at a rate that would let somebody reconstruct, not just where you went, but how long you sat in the parking lot before you went inside. Stay tuned for more disturbing history.

We'll be back after these messages. The litigation is still moving. The consolidated federal case is in the Northern District of Georgia, docketed as a multi-district litigation involving consumer vehicle driving data tracking. In 2006, the court allowed claims under the Federal Wiretap Act and the Stored Communications Act to proceed, along with claims for unjust enrichment and portions of the plaintiff's invasion of privacy, civil conspiracy, and Fair Credit Reporting Act claims. And Hill, the reporter, did something I appreciated as an investigator.

She went through the enrollment process herself to see what a customer would actually experience. There's a second half to the General Motors story that got much less attention, and I think it's the more important half. In April of 2024, Senators Ron Wyden of Oregon and Edward Markey of Massachusetts sent a letter to the Federal Trade Commission asking for an investigation into eight automakers, not for selling data to insurers, for lying about what they do when police came asking. Here's the background. The auto industry, through its trade association, had adopted a set of voluntary privacy principles years earlier.

Those principles included a commitment that the companies would require a warrant or court order before turning over customer location data to government agencies. Wyden's office asked the association how its members actually handle those requests. Five companies were honoring the pledge. General Motors, Ford, Honda, Stalantis, and Tesla required a warrant, absent an emergency, or customer consent. Ford had adopted that standard recently, after conversations with Wyden's office. Hyundai accepted a warrant or other court order,

which the senators noted was less protective, but still met the industry's own stated bar. Toyota, Nissan, Subaru, Volkswagen, BMW, Mazda, Mercedes-Benz, and Kia, all confirmed they would hand over location data to United States government agencies in response to a subpoena. A subpoena does not require a judge. It does not require probable cause. In many cases, it can be issued by a prosecutor or an agency without any court, reviewing it at all. Volkswagen drew a line of sorts. It said it would require a warrant for more than seven days of location data, but would produce six days or less on a subpoena. I'd love to know how that number was arrived at. Six days of a person's movements is a lot of days. And of all the

companies surveyed, only Tesla told the senators it notifies vehicle owners when the government demands their data. After the Supreme Court's 2018 decision in Carpenter v. United States, Obtaining historical cell site location records from your phone company generally requires a warrant. Your phone got constitutional protection. Your car, according to eight major manufacturers, did not. The vehicle you drive to work every morning was, for years, an easier target than the phone in your pocket. And nobody told you. I want to slow down here, because there's a category of harm in this story that has nothing to do with police at all,

and it's the one that kept me up. In December of 2023, the New York Times reported on a Louisiana woman named Christine Dowdall. She had left an abusive husband. She was driving a Mercedes-Benz. Her ex-husband tracked her through the manufacturer's connected vehicle app. He used it to find her. He used it to harass her, and to harass the people who were giving her a place to stay. She had a restraining order.

She had documentation. A sheriff's office in Bosse A barish contacted Mercedes. on her behalf. The company's position, as reported, was that it could not block his access because his name was on the title. Reuters published a similar account around the same time involving a woman being tracked and harassed through Tesla's app despite a restraining order. Think about what the connected car app actually does. It shows real-time location. It shows where the vehicle has been. It can lock and unlock the doors, start and stop the engine, control the climate, and honk the horn. In some vehicles it can access cameras. For a woman fleeing a violent partner, the car is the

escape route. It's also in these cases, the leash. Advocates who work with survivors described running into the same wall repeatedly. Customer service representatives told them that because the abuser was a co-owner, nothing could be done. Divorced decrees didn't matter. Restraining orders didn't matter. Court orders granting sole possession of the vehicle didn't matter. A Queen's attorney who represents survivors made a point that stuck with me, which is that these women are trying to get themselves and their children somewhere safe, and the process being offered to them requires days of phone calls and emails and follow-ups. The Federal Communications Commission sent letters to nine of the largest automakers in January of 24, asking what safeguards they had.

Roman Jessica Rosenwarkel said no survivor of domestic violence and abuse should have to choose between giving up their car and allowing themselves to be stalked and harmed by those who can access its connectivity and data. The agency filed a notice of proposed rulemaking on the 8th of April that year. In March of 2025, Representatives Dan Crenshaw of Texas and Debbie Dingell of Michigan introduced a bipartisan bill called the Safe Vehicle Access for Survivors Act, which would establish a process for survivors to request the connected vehicle services be terminated or disabled. California went ahead and passed its own law requiring manufacturers to set up a clear process, to accept a restraining order or similar documentation,

and to terminate another driver's remote access within two business days. It also requires that drivers be able to shut off location access from inside the vehicle. No carmaker officially opposed it. which raises a question I can't answer in which nobody has answered for me. If none of them opposed the law and all of them agreed the goal was worthy, why did it take a federal agency, two acts of Congress, a state legislature, and a series of newspaper investigations before a woman with a restraining order

could get her ex-husband's phone to stop showing him where she parked. The technology to do it existed the entire time. It's a permission setting. It's a line in a database. So that's the car. Now let's talk about the road. The technology that reads your license plate as you drive past is older than most people assume. And it was invented for exactly the reason you'd guess.

In 1976, the police scientific development branch of the British Home Office built the first automatic number plate recognition system. The purpose was counterterrorism. Working prototypes existed by 1979. The first cameras went into service in 1981 at the Dartford Tunnel. and along the A1 Road, and that year produced the first arrest ever credited to the technology for a stolen vehicle. Then it sat there for a decade, because the cameras were expensive and the software was bad. What changed everything was the provisional Irish Republican Army bombing campaign in the city of London. In 1993, British authorities threw accordion of cameras around the financial district and called it the Ring of Steel.

Every vehicle entering or leaving the square mile got photographed and its plate red. It was the largest deployment of its kind in the world at that point. The pattern set there is worth noticing because it repeats. A specific, severe, identifiable threat justifies a surveillance perimeter. The perimeter works, or at least it doesn't fail conspicuously. The threat recedes. The perimeter stays. By 2002, the home office was running a multi-fellant,

phase program called Project Laser across nine police forces, testing dedicated intercept teams built around plate reading. It worked well enough that the police standards unit put 32 million pounds into a national automatic number plate recognition data center, a national repository of vehicle sightings. In 2003, London launched its congestion charge using roughly 700 plate reading cameras on every road entering and leaving the charging zone. agencies started adopting the technology around 1998, and for a long time, it stayed expensive and specialized. The units cost $20,000 or more apiece and often needed hard wiring. You'd see them

bolted to the trunk of a cruiser, two big boxes, scanning parking lots. I actually drove one of these cruisers when I was a cop in Atlanta. In 2013, a man who goes by the handle, Puking Monkey, gave a presentation at the DefCon Hacking Conference. He'd spotted one of those camera equipped police cars in New Jersey and started thinking about all the ways a vehicle can be tracked. He got interested in his EZ pass, the little radio transponder that lets you pay tolls without stopping. He rewired it so that it would light up and make a noise every time it was red. Then he drove around New York City. The thing went off constantly, not at toll plazas, on surface streets in Midtown, at 42nd and 8th. On his way through the Lincoln Tunnel in a spot with no toll

booth. Kashmir Hill, the same reporter who would later break the General Motors story, wrote it up for Forbes. The New York City Department of Transportation confirmed the readers existed. A spokesman said they were on highways across the city and on streets in Manhattan, Brooklyn and Staten Island, and had been in use for years. They were part of a traffic management program called Midtown in Motion. The vendor said the transponder identification is scrambled to make it anonymous, held in memory for several minutes to compare against sightings from other readers and used to calculate travel times. That may well be exactly true. I have no evidence it isn't. Here's what bothered Hill, and what bothers me. The department wouldn't say precisely what was read

or how long the geolocation information was kept. And the fact that the transponders would be read outside of toll collection did not appear anywhere in the terms and conditions people agreed to when they got one. The New York Civil Liberties Union filed a public records request to find out. In early 2017, a Georgia Tech electrical engineer named Garrett Langley had a break-in problem in his Atlanta neighborhood. He talked to a sergeant about it. What he took from that conversation was that a license plate number is usually the thing

that cracks a property crime case, and that most departments couldn't afford to blanket a county with plate readers at $20,000 a poll. He and a fellow Georgia Tech graduate named Matt Fury, along with Paige Todd, launched flock safety on the 1st of March, 2017. The pitch was evidence, not footage. The product was a small solar-powered camera with a cellular connection that you could strap to a pole in an afternoon, no trenching, no electrician. The first customers weren't police departments.

They were homeowners associations and private neighborhoods. The company went through Wycombe, the following year, and then it grew in a way that I don't think anybody involved fully anticipated. As of the reporting available this year, Flock operates in more than 5,000 communities across 49 states. The American Civil Liberties Union puts the number of cameras at more than 120,000 in at least 6,000 cities. The company's own figures describe processing something on the order of 20 billion vehicle reads a month. Its valuation was reported at 8.4 billion. in April of 2006, on annual recurring revenue north of 300 million, with close to a billion

dollars raised and roughly 1,500 to 1,600 employees. In under a decade, a startup that began by selling cameras to a neighborhood association became the operator of what is, functionally, a national vehicle movement database, not a government one, a private one that governments rent access to. Let me explain what the camera. actually does because the phrase license plate reader undersells it badly. The camera photographs every vehicle that passes, not vehicles on a wanted list. Every vehicle. It reads the plate using machine learning, time stamps it, and logs the location. That much you'd expect, but the system doesn't need the plate. Flock builds what it calls a vehicle fingerprint. The

software catalogs make, model, body style, color, and distinguishing features. roof racks, bike racks, bumper stickers, damage, a ladder in the bed, a dent in the quarter panel, which means investigators can search for a car when they have no plate number at all. Show me every white pickup with a toolbox and a cracked taillight that crossed this intersection between two and four in the morning. That's genuinely powerful. I'd have loved it. It also means that covering your plate accomplishes nothing,

and that a system built to identify vehicles has, in practice, been taught to identify a specific vehicle belonging to a specific person by the accumulation of small details that make it unlike any other car on the road. Default retention has historically been 30 days. Some states have written shorter windows into law. Virginia caps it at 21 days. Washington and Connecticut have moved in the same direction. 30 days doesn't sound like much until you consider that 30, days of your movements includes every place you slept, every doctor you visited, every meeting

you attended, every church service, every bar, every union hall, every A.A. meeting, and every night you didn't go home. Norfolk's police chief, Mark Talbot, described the goal of his city's deployment in a way I found unusually candid. He wanted a nice curtain of technology. He wanted it to be difficult to drive anywhere of any distance without running into a a camera somewhere. He got it, and two of his residents sued him over it, which we'll come back to. In November of 2025, police used flock cameras to help track down the suspect in the mass shooting at Brown University. In August of 2006, Gaught police in California said the cameras helped them locate a kidnapping suspect within five minutes of being alerted by another agency

that a man was driving a white Suzuki with a victim in it. They arrested Everett Master, on kidnapping and false imprisonment charges among others. The Milwaukee Police Department provided CNN with summaries of 24 cases in April and 26 more in May where plate readers played a role. In one, they caught a suspect accused of domestic violence, vehicle theft, and firearm possession. Cleveland's Department of Public Safety

compared case closure rates for robbery, felonious assault, assault and burglary, before and after the cameras went in. The combined closure rate rose from 15.5% to 18.7%. Flock's chief executive says the company supported roughly a million law enforcement investigations last year, and helped locate more than 10,000 missing persons. He told CNN that when he travels the country, he hears a long list of success stories that would have been cold cases without the technology.

A 3.2 percentage point improvement in closure rates is a real improvement. It is not nothing. It's also not a revolution, and it's the kind of number that can move for a dozen reasons unrelated to cameras. Cleveland's own officials said that producing a definitive count of crime solved because of the technology would require reviewing thousands of individual investigative files

to determine whether the plate reader generated the initial lead, corroborated other evidence, or played almost no role at all. They excluded stolen vehicle cases entirely, because a surge in Kia thefts during the study window would have skewed everything. Thaddeus Johnson, a former law enforcement officer now at the Council on Criminal Justice, put it about as well as it can be put. The cameras are excellent at generating leads. Leads are not solutions,

and supported an investigation is doing an enormous amount of work in that million case figure. I've written a hundred supplemental reports where I checked a database, got nothing useful, and documented that I checked it. That's a system being used. It isn't a case being solved. So here's where I land, and I want to be clear about it before we go into the dark part of this story. The tool works. The tool is useful.

A kidnapping victim came home in five minutes because of it. None of that is an argument about who else gets searched. How often? By whom? And with what supervision? Those are separate questions, and the industry has spent nine years answering the first one loudly. and the second one, not at all. The Institute for Justice, a public interest law firm,

maintains a database of documented abuses of automated license plate reader systems. As of the 12th of August, 2006, it lists 147 incidents. They built it from news reports and public records. They exclude vehicle-mounted units, speed cameras, red-light cameras, and toll cameras, and count only stationary, always-on systems.

They organized the incidents into four categories. Stalking. Error. Non-law enforcement use. Other misuse. I read the entire thing. It took most of an evening. I'm going to walk you through some of it. And I'm going to use names because these are public records and public arrests.

And the people involved were paid by the public. We'll start in Kansas. In October of 2022, a lieutenant in Kichi named Victor Hayar used flock cameras to track his estranged wife. He pleaded guilty to computer crime and stalking. A year later, about 20 minutes up the road in the town of Sedgwick, the police chief did the same thing. Lee Nygaard searched for his ex-girlfriend's vehicle 164 times over several months and searched for her new boyfriend's vehicle 64 more. He resigned. Two police officers, two small departments, same state, same tool. Within about a year of each other. Now go east to

Georgia, which has produced more documented cases than anywhere in the country. In July of 2024, the police chief of Brazzleton, Michael Steffman, was arrested for allegedly using plate readers to stalk and harass multiple people, including a former romantic partner. He resigned shortly before the arrest. In Richmond County, a deputy named Jacarius Yarbrough was arrested after allegedly running the plate of a woman he'd been having an affair with, 1,630,000. times over several months. 1,639 searches. Over a few months.

That's a man sitting down and typing her plate into a surveillance system somewhere between 15 and 20 times a day. Every day for months. To find out where she was. The following month, three more Richmond County Sheriff's officers were arrested for allegedly misusing the system more than 100 times each.

Stay tuned for more disturbing history. We'll be back after these months. messages. Sources who spoke with the Augusta Press said some of those searches involved tracking women. In June of 2026, Cherokee County fired and jailed a lieutenant and a sergeant after an internal audit. In July, the Albany Police Department fired five officers and had them arrested. A DeKalb County Sergeant was fired and arrested. Three Fayetteville officers were fired and placed under criminal investigation. Four Savannah officers and two civilian staffers were put on leave and referred for criminal investigation. A Habersham County deputy, a Green County deputy,

a major in Millageville, a Henry County officer. In August, three deputies in Bibb County were charged. Three Coweta County employees resigned in the middle of an audit. Five D-Cab officers were suspended for searching without a legitimate law enforcement purpose, and three more were suspended for unspecified violations. That's one state. over roughly 90 days. Elsewhere in the country, the same pattern repeats with different names on it. Haynes City, Florida. Officer Christopher Goodson searched the database for his estranged wife's car more than 700 times between September of 2024 and June of 26. Louisville, Kentucky. Officer Roberto Sedino was charged with multiple felonies after allegedly tracking an ex-partner

and her friends hundreds of times over two months. Milwaukee. Officer Josuea Ayala allegedly tracked a woman he was dating and her ex-partner close to 180 times in two months. He resigned after being charged with misconduct in public office. And here's the detail from Milwaukee that I've turned over in my head more than any other single fact in this episode. The detective assigned to investigate Ayala's misconduct

was himself later charged with misusing the same. system, allegedly tracking two people over the course of a year. The Internal Affairs Investigation found the abuse. Then the investigator turned out to be doing it too. Jerome County, Idaho. Sheriff George Optuck used the system to search for his wife's vehicle hundreds of times. The Idaho Attorney General concluded no crime had been committed. Optic retired in April of 2006, two years before his term ended. Key West, Florida. Deputy Lamar Roman met a woman while working security on a television set. He used a plate reader system to track her, then pulled her over.

He was arrested and charged with accessing a computer or electronic device without authorization. In his interview with investigators, according to reporting by 404 media, he said he didn't need to do it, and that he'd known right when he did it. About the traffic stop, he said, yeah, I know it's stupid. She wasn't an ex. She wasn't a suspect. She was a stranger who happened to be standing near him one day, and the state of Florida had given him the ability to find out where she lived. There's a case in Georgia that the Institute for Justice highlights near the top of its database,

citing Washington Post reporting from August of this year. A police chief tracked his ex-girlfriend and her teenage daughter roughly 600 times. Now, the standard institutional response to all of this is that these are individual bad actors, that no system is immune to misconduct, and that audits catch it. That isn't a stupid argument. Every agency I ever worked for had somebody in it who shouldn't have been there. The existence of a corrupt officer doesn't indict a tool. But two facts from the record undercut it.

The first is how these cases surface. The Institute for Justice reviewed how each incident came to light. Only a small handful were initially caught by internal investigations. Most of them surfaced because the victim went to the police, usually in the course of reporting stalking. And the plate reader misuse came out during that, which means the audit didn't catch it. The woman being hunted caught it and then had to walk into a police station and explained to police that a police officer was using police equipment to find her. The second fact comes out of the Norfolk litigation, and it's the single most damning thing I read in the course of preparing this. The Institute for Justice in Discovery Against the City of Norfolk obtained the search logs.

Over a two-year period, there were roughly 230,000 searches of the system by Norfolk officers. Every one of those searches required the officer to type a reason. The top three reasons entered, accounting for nearly a third of all searches, were CJ, stolen, and INV. Two of those aren't reasons. And somewhere in that log, one officer repeatedly, typed la la la la nobody noticed nobody flagged it nobody called him in he typed nonsense into an accountability field over and over ran his searches and went about his day and the only reason anybody in the world knows about it is that a non-profit law firm sued the city and got the records in discovery

the american civil liberties union of massachusetts found the same thing at the network level When Flock rolled out a tool intended to flag improper searches by analyzing the reason field, the organization's review of the audits found officers routinely entering vague terms like investigation, or S-U-S-P. An accountability system that depends on the honesty of the person being held accountable is not an accountability system. It's a suggestion box. Then there's the second failure mode, which has nothing to do with anyone's motives. The cameras are wrong a lot. In flock states under optimal conditions, its cameras read more than 96% of license plate characters accurately.

That sounds pretty good, right? Then Business Insider obtained hundreds of pages of records from the Roseville Police Department in California through public records requests. Between 2003 and 2024, the system sent Roseville Police 1,427 alerts, flagging vehicles as stolen or connected to a felony. The department's own analysis found that in 1,011 of those alerts, 71% the software had misread the plate. The errors were mundane. A nine red as an eight. A three red as a two.

A one red as a four. One resident's plate got misread at least six times, to the point that a dispatch supervisor wrote that it had gotten easier because they'd memorized it. An officer eventually called the man to explain why his car kept coming up stolen. and he offered to take the cover off his plate. In April of 2025, a supervisor in Roseville's Real Time Crime Center told the company that 16 of the 17 plates flagged the previous week had been wrong. I want to be precise about what that 71% means

because it's been reported carelessly in places and I don't want to be one of them. It does not mean the cameras misrid 71% of all plates. It means that among the alerts, the small subsubstances, set of reads that actually triggered a notification to police, 71% were wrong. A system can read plates correctly, the overwhelming majority of the time, and still generate mostly false alarms, simply because genuinely wanted vehicles are rare. Roseville's spokesman made that clarification himself, and it's a fair one. Roseville also says that none of the bad alerts led to a stop or an arrest, because their

officers verify a plate visually before acting. The company said the misread stem from a non-standard deployment, older hardware, cameras mounted farther from vehicles and higher than recommended, and a request to capture only the backs of vehicles. It said performance has since improved significantly. The Roseville Police Department disputes that. Max Isaacs, who directs technology law and policy at New York University's policing project, called the rate stunning and said that in the absence of concrete rules addressing misreads. The technology isn't safe to deploy in communities, Morristown, Tennessee, June of 2024. A camera read the letter O as the numeral zero. Officers detained two grandparents at gunpoint while their three-year-old granddaughter watched from inside the car,

Toledo, Ohio, two months earlier. A camera read a seven as a two. Officers detained an innocent driver at gunpoint, released a police dog on him, and held him in jail for several hours. Española, New Mexico. A misread digit put two sisters on the pavement at gunpoint. Aurora, Colorado, August of 2020. A misread on a minivan ended with a woman and four children, detained at gunpoint. Detroit. Officers misinterpreted the data, detained a woman at gunpoint, put her autistic child in the back

of a patrol car and impounded her vehicle for weeks. The city later settled, Lafayette, Indiana. A man detained at gunpoint in a Chick-fil-A drive-thru because a camera got his plate wrong. Volusia County, Florida, April of 2006. Officers matched an innocent driver's car to one that had been at the scene of a fatal crash. She spent 13 days in jail before the charges were dropped. San Diego, November of 2025. A capture linked a man to a violent crime. He was five miles away when it happened. He spent nearly a month in jail.

Then there's a whole separate category where the camera worked perfectly, and the humans failed. Hot lists that nobody updated. In Hercules, California in 2018, a rental car was reported stolen, then recovered, and nobody took it off the list. So a reader flagged it, and a man and his brother ended up at gunpoint.

Same story in Greenville, South Carolina, with two women. Same story in Jackson Township, Ohio. And in Colorado, over roughly eight months across three separate counties, three different drivers were repeatedly pulled over because officers had put their plate numbers on a hot list by mistake. Not once. Repeatedly. Here's the thing about all of this that I don't think gets said clearly enough.

A false alert from a plate reader isn't like a false positive on a spam filter. The output of the system is armed police officers approaching a vehicle believing it was used in a felony. That's the most dangerous 90 seconds in American law enforcement, even when everybody has good information. Feeding it bad information at scale isn't a software problem. It's a use of force problem wearing a software costume. Everything I've described so far could theoretically be handled locally. A department buys cameras, a department misuses cameras, a department gets sued. That isn't the architecture.

The reason this became a national story is a feature called National Lookup and the reciprocal sharing model behind it. When an agency joins, it contributes its camera data to the pool and gains the ability to search everybody else's. The more agencies join, the more valuable membership becomes, and the harder it becomes to opt out without losing action. access to everyone else. Senate records reported in August of 2025 indicated that about 75% of the company's law enforcement customers had enrolled in the national database. The American Civil Liberties Union of Massachusetts found that in a single 30-day window in the spring of 2025, more than 450,000 searches hit the national database. The Electronic Frontier Foundation obtained something larger. Datasets

representing more than 12 million searches logged by more than 3,900 agencies between December of 2024 and October of 2025. 12 million searches, 10 months. And once you have that data, you can ask what people were actually searching for. Start with immigration, because that's where the audits started firing. 404 media reported that the lookup tool had been used for at least 4,000 immigration-related searches between the presidential inauguration and May of 2025. In some cases, that meant federal immigration authorities reaching into data from states whose laws expressly prohibit local police

from collaborating with them. In Illinois, Secretary of State Alexei Giannulius ordered an audit in June of 2025. Illinois had passed a law in 2003, building on an earlier statute called the Tamara Clayton Expressway Camera Act that specifically prohibited plate reader data from being shared for enforcing immigration law or out-of-state abortion investigations. The audit sampled 12 local agencies. On the 25th of August, his office announced that United States Customs and Border Protection had gained access to Illinois license plate camera data, a violation of state law. The audit found the company did not have proper safeguards in place for data sharing, and that this was compounded by a pilot program with customs and border protection,

that according to the state's account, the company's own leadership was unaware of. A surveillance company running a federal pilot program that its own executives allegedly didn't know about. The company's chief executive posted a response acknowledging that they'd communicated poorly and that they hadn't created distinct permissions and policies. He said the company had no contractual relationship with any Department of Homeland Security Agency. but had engaged in limited pilots with customs and border protection and homeland security investigations to assist with human trafficking and fentanyl distribution. Illinois ordered federal access shut off.

The pilots were paused nationwide. Ten days later, the Secretary of State's office found a second violation, this one involving a different vendor, Motorola Solutions, and the Village of Forest Park Police Department, again giving customs and border protection access to cameras. data. Meanwhile in the suburb of Mount Prospect, the state said there had been 262 immigration-related searches between the 14th of January and the 30th of April. When the Texas abortion search first surfaced, the Mount Prospect chief said he hadn't known his department had even opted into the national

feature. This kept happening everywhere. In Dayton, Ohio, records showed more than 140 law enforcement agencies nationwide access the city's camera system for immigration related searches over several years, despite city policy not permitting it. In San Francisco, out-of-state authorities ran more than 1.6 million searches of the city's plate reader data, including searches for immigration enforcement purposes that are illegal under California law. In Egan, Minnesota, out-of-state agencies were able to search twin cities suburban data for civil immigration enforcement, despite a state law restricting that data to criminal matters.

In Richmond, Virginia, an analyst with the Bureau of Alcohol, Tobacco, Firearms, and Explosives, accessed the city's network for immigration-related queries, despite an explicit city policy barring that kind of federal cooperation. Separately, a Richmond sergeant was found to have violated the same policy by sharing data with the Federal Bureau of Investigation. In Louisville, a detective was disciplined for sharing his login with a drug enforcement administration agent, who then used it to run immigration-related search terms. In Sparks, Nevada, the Electronic Frontier Foundation found the department was receiving immigration violation, hot list alerts from immigration and customs enforcement, despite an internal policy prohibiting exactly that.

What I'm describing here isn't a policy disagreement. It's a structural fact. Cities and states wrote rules about what their own surveillance data could be used for. Then the data left, through a shared name. network into the hands of agencies that had never agreed to those rules and weren't bound by them. In several documented cases, the local officials didn't know it was happening until an auditor or a reporter told them. Whatever you think should be done about immigration, a village police chief in Illinois doesn't know who's searching his cameras is not a defensible way to run anything.

Now, the part that I think should concern people across the political spectrum, and historically hasn't, because everybody assumes it'll only ever be pointed at the other side. The Electronic Frontier Foundation went through those 12 million search records looking for protest activity. More than 50 federal, state, and local agencies ran hundreds of searches connected to political demonstrations. The organized protests of February 2025. The hands-off demonstrations in April.

The No King's demonstrations in June and October. 19 agencies ran searches specifically tied to the No King's events. In Lubbock, Texas, the police department ran two searches on the 15th of March for a Tennessee license plate that corresponded to a rally about the mental health impact of immigration policy. The logged reason was protest vehicle. Those two searches reached 5,96 networks. In Grand Rapids, Michigan, five searches connected to a February rally. reason logged as protest, reaching roughly 650 networks.

In Spokane, a sheriff's office tactical analyst ran two searches on the 17th of February, the day of nationwide demonstrations, with the reason SPD protest. The first swept 470 cameras across nine networks. The second swept 1,29 cameras across 74 networks. No arrests were made in connection with that protest. no explanation was given in the log for why the searches were necessary. Three agencies used the system to target activists from an animal rights organization that uses civil disobedience against factory farms.

Delaware State Police queried the network nine times in March of 2025 in connection with those actions. In Akron, Ohio, a police analyst and an officer tracked a prominent activist who had publicly criticized the department. The reason they typed into the box was activist recording that violated their own department's policy. In Lenoxa, Kansas, police allegedly used plate readers to track the movements of a writer who had published a column critical of them. A cop looking up a woman he's obsessed with is a crime, and it's a personal failure, and the man goes to jail. That's terrible, and it's also, in a certain sense, ordinary. Men have abused access to information about women since there was information to abuse. An agency searching a national surveillance network for the license plates of people who attended a lawful demonstration is a different species of thing.

Nobody's obsessed. Nobody's in love. It's just Tuesday at the office. And somebody types protest into a box. And a machine returns the movement history of citizens whose only act was showing up somewhere and holding a sign. There's not much daylight between that and writing down the plates outside of church. And in the Akron case, in the Lenoxa case, the person on the receiving end was somebody who had publicly criticized the police. That's not surveillance in service of an investigation. That's the oldest instinct in institutional life, which is to find out things about the people making your life difficult. Stay tuned for more disturbing history. We'll be back after these messages.

There's one more product I need to tell you about. and then we'll get to the courts. In May of 2025, 404 media obtained internal presentation slides, Slack messages, and meeting audio describing a new flock product called Nova. The reporting described a tool designed in the words of the internal material

to let police jump from a license plate to a person. Nova was built to combine plate reader data with public records, commercially available data from brokers and law enforcement databases. so that an officer could go from a car passing a camera to a name, an address, and an associated network of people, quickly, without a warrant or court order. The reporting also said the data sources under discussion included information obtained from data breaches, including a 2,021 breach of a parking payment app.

The internal reaction is what made it a story. One employee wrote, according to the messages obtained, that they were pretty horrified to hear the company used stolen data in its system, and that beyond having been obtained illegally, it seemed like it could create perverse incentives for more data to be leaked and stolen. That's a company employee in a company channel, raising the point that building a business on hacked data creates a market for hacking. On the 30th of May, the company held an all-hands meeting, and then published a blog post titled, Correcting the Record. It said the policy decision had been made that flock would not supply dark web data, meaning Nova would not supply any data purchased

from known breaches or stolen data. A spokesperson said the original story had been published prematurely, while the team was still deciding what sources the product would use, and that the reporting had relied on rumors and early product discussions. A company statement also argued that officers may already have access to similar information through other means, and that centralizing it inside their platform adds a layer of transparency and accountability so that elected bodies can ensure it's used lawfully. That last argument is more interesting than it sounds, and I don't want to dismiss it out of hand. There's a real case that a single audited system is easier to oversee than a dozen unaudited ones. The trouble is that the oversight has to actually happen.

And we already know what the reason field looks like in practice. It looks like INV. It looks like La La La La. A security researcher published an analysis in which he says the Nova front-end code defines, fetches, stores, and displays a data source explicitly labeled dark data, with search selectors for social security numbers, credit card numbers, cryptocurrency wallets, and similar identifiers. I have not independently verified that analysis. It comes from a single source outside traditional journalism, and the company denies the underlying claim. Let's go to the courts, because this is where the story is genuinely unresolved. In October of 2004, the Institute for Justice filed a federal lawsuit against the city of Norfolk,

its police department, and Chief Mark Talbot, on behalf of two residents. Lee Schmidt is a Navy veteran who lives in Norfolk. His objection was specific and modest. He didn't want the city keeping a record of his trips to his daughter's schools or to the shooting range. Crystal Arrington lives in Portsmouth and works as a home health aide visiting patients in Norfolk. Her concern extended to her patients because a record of where a home health worker goes is also a record of who is sick and where they live. Norfolk had installed somewhere between 172 and 176 camera clusters starting in 2003. The city retains data for 21 days.

The state maximum. Officers can query it without a warrant. Over four months in 2025, the system captured Schmidt's vehicle 475 times. It captured Arrington's 325 times. That's Schmidt getting photographed by his own city roughly four times a day, every day, for a third of a year, for driving. In February of 2025, Chief Judge Mark Davis denied the city's motion to dismiss. He leaned on Carpenter, the 2018 Supreme Court decision requiring a warrant for historical cell phone location records. He wrote that a reasonable person could believe society's expectations,

as the Supreme Court laid them out in Carpenter, were being violated by the Norfolk system, and that the complaint alleged facts notably similar to those the court had found to clearly violate privacy expectations. He described what Carpenter condemned as law enforcement secretly monitoring and cataloging the whole of tens of thousands of individuals' movements over an extended period. Flock itself tried to intervene in the case in the middle of 2025. Judge Davis denied that, too. In an opinion noting the company had made a conscious gamble not to show up on time, that it wasn't the court's fault the train had already left the station,

and that letting them in at that point would throw the case off the rails. And then, on the 27th of January, 2006, the same day, judge granted summary judgment for the city. His 51-page opinion held that the plaintiffs could not demonstrate the system was capable of tracking the whole of a person's movements. 176 cameras with a rolling 21-day window, he reasoned, wasn't the exhaustive surveillance the Supreme Court had worried about in Carpenter, and wasn't comparable to the constant aerial surveillance the Fourth Circuit had struck down in the Baltimore case. But he didn't write it as a clean win. He wrote that rapid technological advances make it impossible to predict how police surveillance will evolve,

and that Fourth Amendment analysis has to stay nimble. He wrote that this kind of surveillance could become too intrusive and run afoul of constitutional standards at some point. And then he wrote the line that everybody quoted, while a definitive answer to that question is elusive. What is readily apparent to this court is that, at least in Norfolk, Virginia, the answer is not today. Schmidt and Arrington appealed to the Fourth Circuit, where the case is pending, and then five months later, the ground moved underneath everybody. On the 29th of June, 2006, the Supreme Court decided Chattree v. United States. The case came out of a 2019 credit union robbery in Virginia, where investigators used a geofence warrant to make Google produce location history for every device in an area around the crime scene during a window of time. The court held, six to three, that police conduct a Fourth Amendment search when they obtain a person's location data that way.

Justice Elena Kagan wrote for five of the justices, joined by the Chief Justice. She wrote that an individual has a reasonable expectation of privacy in records about his cell phone's location, and that police intrude on that constitutionally protected interest when they demand the information, even for a limited time, and even from a third-party technology company. The court sent the case back down without deciding whether that particular warrant was valid. Two pieces of that reasoning matter enormously for the road cameras. The first is that the court rejected the argument that voluntarily letting a company collect your location strips away your privacy interest in it.

That's the third-party doctrine, and the court declined to let it swallow location data. The second is more direct. The government argued that even if the fourth-party doctrine, the court declined to let it swallow location data. Amendment applied, the search here was fine because it pulled only a narrow, time-limited slice of a much larger data set. The court rejected that, too. Once the Fourth Amendment applies, the majority reason, it doesn't matter how small a bite investigators took out of an all-encompassing database. That second holding aimed squarely at the argument Norfolk won on. The city's defense was, in essence, 21 days is a small slice. I'm not a lawyer, and I'm not a lawyer,

I'm not going to predict what the Fourth Circuit does with it. But I've read enough appellate opinions to recognize when a district court's reasoning has been undercut by a decision handed down after the ruling. And this looks like one of those. I want to come back to Texas now, because I opened with it, and because it's the case where every thread in this episode ties together into one knot. On the 9th of May, 2025, deputies with the Johnson County Sheriff's Office in Texas responded to a home. A woman had self-administered an abortion. According to the reporting, the incident itself had happened more than two weeks earlier. Deputies treated the residents as a potential crime scene. They filled out a crime scene log. They put up crime scene tape. They photographed a

stainless steel tumbler that the woman's partner told them contained the fetus and placed it in a small body bag. And according to the report, while they were at that house, deputies discussed options for criminal charges related to the abortion. That same day, a deputy ran the search of the National Plate Reader Network. The reason he typed was, had an abortion, search for female. The search reached more than 83,000 cameras across 6,809 law enforcement networks, including networks in Washington and Illinois, where what she had done was legal. Both queries carried the same case number as the sheriff's office's formal incident report. 404 media broke the story in May of 2025.

The public response from officials was immediate and specific. Sheriff Adam King told 404 media that her family was worried she was going to bleed to death and that they were trying to find her to get her to a hospital. He later told the Dallas Morning News that they were just trying to check on her welfare and get her to a doctor if needed or to the hospital. The company called the Reporting False and Clickbait and its chief executive characterized it as a misleading account by an activist journalist. That's a valid story.

A missing person, a worried family, a department using a tool to save a life. If that's what happened, this whole episode has a very different shape. So the Electronic Frontier Foundation went and got the court records. The affidavit written by the detective who was actually on the scene makes no mention of concerned family members and no mention of a medical investigator. A supplemental report claiming that officers had found a large amount of blood at the scene was filed by a desk sergeant a month after the incident and after the story had already broken. The on-scene detective's own affidavit doesn't mention it.

The documents show deputies discussing potential criminal charges with prosecutors. On the same day, the search was run, and the timeline doesn't work. The death investigation that supposedly triggered the search began roughly two weeks, after the abortion, which makes the fear that she would bleed to death an odd thing to be responding to in real time. I've written affidavits. I've written supplementals. And I'll tell you what a supplemental report that appears a month later contradicting the on-scene officer's account and adding a detail that conveniently justifies a search that just made national news. Looks like to somebody who spent time in that world. It simply looks like paper. I want to be clear that I can't

prove that. It's possible the desk sergeant genuinely learned something the detective didn't record. But when the story an agency tells the public is contradicted by the sworn account of its own detective, the burden shifts. And it doesn't shift back because somebody calls a reporter a name. The consequences of that one search reached a long way. Illinois cut off out-of-state access to its cameras in June of 2025, then launched the audit that found the federal violations. Senator Senator Ron Wyden negotiated an agreement to give Oregon drivers the same protective filters, effective the 25th of July 2025, after confirming with the Oregon Attorney General and State Police that no statewide policy existed. Congressional ranking members Rajah Krishna Mordi

and Robert Garcia opened an investigation, calling the search a gross misuse and abuse of surveillance technology, and demanding a full accounting of every national lookup search involving the terms abortion, ICE or CBP. They pointed out something I think is the heart of it, which is that individuals cannot choose whether their license plate data is collected and scanned, and that once it is, it becomes accessible to officials from unrelated jurisdictions across the country. Kate Bertash of the Digital Defense Fund made the point that anti-abortion groups have a documented history of recording license plates outside clinics, and that a system permitting tracking over time and across state lines undermines protection even for people who

live in states where the procedure is legal. That's the mechanism that should worry you no matter where you stand on abortion itself. A camera in Springfield, Illinois, was installed by people who wrote a law saying it could not be used this way. It was used this way anyway, by an agency two states over, because the camera was plugged into a network, and the network doesn't care about the Illinois legislature. The reckoning arrived in 2006, and it came from directions the industry probably didn't expect. Cities started canceling. Austin, Texas, Cambridge, Massachusetts, Flagstaff, Arizona, Oak Park, Illinois, Chandler, Arizona, after an employee allegedly misused the data. Dozens of agencies nationwide have ended their contracts, citing privacy concerns, federal access to local

data and the documented abuse cases. In Santa Cruz, an official reviewing audits found what appeared to be immigration-related searches by California police departments and called it very concerning. When the company introduced keyword filters to block searches referencing civil immigration or reproductive health care where state law forbids it, and then added a drop-down menu requiring officers to pick an offense type, she wasn't impressed. Her concern was obvious and Correct. An officer running an immigration search can simply select a more palatable option from the menu. In February of 2006, Ring ended its partnership with the company after public backlash. In April, there were protests in Atlanta and on college campuses, and people started taking the

cameras down themselves, physically. Reporting indicates protesters have dismantled cameras in at least five states. I'm not endorsing that, and I want to say that plainly. Taring down, municipal equipment is a crime. It puts the person doing it at risk and it doesn't fix anything structural. But I'd point out that you don't get a wave of people climbing poles with tools unless a lot of them have concluded that the ordinary channels aren't working. There's one incident in the database that I keep thinking about because of how small and strange it is. In August of 2006, authorities in Mill Creek, Utah found a flock camera installed on city property. They removed it because they could not determine who had installed it or who had

access to what it recorded. A surveillance camera on public land and the government that owns the land couldn't figure out whose it was. On the 13th of August, 2006, one day before I sat down to record this, the company announced a set of changes. Every law enforcement customer will be required to implement an audit tool intended to flag abnormal search behavior, with users locked out pending internal review when it triggers, mandatory by the 1st of January. Customers will be able to decide which offense types outside agencies can search their data for, which would let a city block immigration-related searches of its cameras. The chief executive said the change gives department's control to use the system consistently, with community values.

Evidence retention windows were shortened. The American Civil Liberties Union said the shorter retention could be a step in the right direction, and characterized the rest as retreads of measures that hadn't worked before. I'd point out one structural fact about that announcement. Every one of those reforms is a company policy, not a statute, not a court order, a policy, adopted by a private business that a private business can amend the next time the pressure eases. Nine years of growth and the accountability framework is still a promise. So where does that leave the rest of us?

I've spent this whole episode moving between two things that are both true, and I don't want to resolve that tension artificially, because I don't think it resolves. A kidnapping victim in Galt, California, came home because a camera saw a white Suzuki. Cleveland's closure rates went up. The Brown University shooting suspect was located. 10,000 missing people by the company's count were found. Those aren't marketing claims I can wave away,

and I'm not going to pretend a career's worth of instinct doesn't tell me those numbers represent real families getting real phone calls. And a deputy in Texas typed a sentence about a woman's medical procedure into a search box and reached 83,000 cameras in a country where nobody voted on that. Both at the same time. Same system. What I've concluded, and you're welcome to land somewhere else, is that the question was never whether the tool,

works. The question is what happens in the enormous space between the cases where it's used well and the cases where it's used badly. And the answer that the record supports is that almost nothing happens because almost nobody is looking. 147 documented incidents and only a handful found by the systems built to find them. 230,000 searches in one mid-sized city over two years and the three most common justifications were C.J., stolen, and NV. 12 million searches nationwide in 10 months, and the way we learned what was in them was that a non-profit filed records requests. I keep coming back to that officer in Norfolk. I don't know his name. I don't know what he was

searching for or whether any of it was legitimate. All I know is that his department built a system that asked him to justify himself, and he typed, La, La, La, La, and ran the search, and the machine gave him what he asked for, and it did that again and again, and nobody in the chain of command ever said a word about it. That's not a scandal about one man. That's a description of how much attention the entire apparatus was actually receiving. And down the road from wherever he was sitting, a Navy veteran named Lee Schmidt was driving his daughters to school and driving out to the range. 475 recorded sightings in four months, in a city whose police chief said out loud that he wanted a curtain of technology thick enough that you couldn't drive any distance without hitting a camera.

Schmidt sued. He lost, and the judge who ruled against him wrote that this kind of surveillance could become unconstitutional at some point, and that at least in Norfolk, at least on that day, the answer was not today. That case is now in front of the Fourth Circuit, argued against the backdrop of a Supreme Court decision that came down five months after the ruling and rejected the exact reasoning the city used to win. I don't know how it comes out. Nobody does. What I do know is that while it's being argued, the cameras haven't stopped. Neither has your car. Right now, in a driveway somewhere, a vehicle with the ignition off is holding a list of every place it's been, every phone that's touched it, and every hard stop its driver has made in the last, however long,

waiting for somebody with the right subpoena to come ask. Nobody handed you a contract for it. there was no I agree button to push. There was just a car and a road, and a camera on a pole and a database, and a text box where a man could type anything he wanted. That's what privacy not included really looks like. Your skin.