AI surveillance systems deployed without adequate oversight and regulation can lead to significant privacy violations, as demonstrated by Flock Safety's network of 90,000 cameras that scans 20 billion license plates monthly, has been used for immigration enforcement despite company denials, and has led cities to cover cameras with bin bags because they cannot determine whether cameras are still recording or how to terminate contracts. The core problem is that regulatory frameworks are inherently reactive, meaning they only address issues after they occur, while AI deployment moves at venture capital speed, creating a fundamental mismatch between technological capability and governance capacity.
Deep Dive
Prerequisite Knowledge
- No data available.
Where to go next
- No data available.
Deep Dive
AI Surveillance Just Crossed A Critical LineAdded:
You were identifiable at every single one of these points and you didn't opt into any of them. Right now in Dayton, Ohio, there are AI-powered surveillance cameras on public streets that are covered in black bin bags because the city government, the actual elected officials responsible for public safety, cannot figure out whether they are allowed to take the cameras down, whether the cameras are even still recording, or how to get out of the contract they signed with the company that installed them. Their solution was to send the police department and the public works team out with bin liners, not as a temporary measure while they figured out the real solution. That was the solution, which to be fair is also my solution when something in the fridge has gone wrong and I don't really want to investigate further.
In Evanston, Illinois, the city terminated its contract with the same company. The company then reinstalled the cameras without permission. That is a level of persistence usually reserved for exes and browser toolbars. The city had to send a cease and desist letter and then, while they waited for the legal process to play out, they also covered the cameras in bin bags. Two different cities, same conclusion, bin bags. These cameras belong to a company called Flock Safety. There are over 90,000 of them across the United States.
They scan 20 billion license plates every single month, which for context is more than the number of registered vehicles in the country, meaning some of you are being scanned multiple times a day and the most exciting thing you did was maybe go to Costco. An growing number of cities are slowly discovering that the surveillance infrastructure they invited in is far harder to get rid of than it was to install. My name is Ella. I have a PhD in computer science and I analyze AI developments to understand what's actually happening beneath all of this hype. In this video, I'm going to explain what Flock Safety is and how it ended up in thousands of cities. Then I'm going to walk through three specific ways the system has already been misused. And after that, I want to talk about something broader, why anonymity in public space is essentially already gone, why most people don't realize it, including in this audience, and what that means for how quickly regulation needs to catch up. Spoiler, not quickly enough, but you probably guessed that from the bin bags.
Flock Safety was founded back in 2017 in Atlanta. The origin story is very clear.
The founder experienced a property crime. The police couldn't help because there was no evidence. So, he built a solar-powered camera that reads license plates automatically and uploads the data to the cloud. Simple, definitely useful. The kind of technology that genuinely does help recover stolen vehicles, find missing persons, and track suspects in violent crime cases.
Nobody serious really is arguing that the core technology has no value.
Somebody gets their car stolen, the camera catches the plate, the police recover it, great. The problem is that nobody has ever raised a billion dollars to just stop at great. The pitch to cities was of course straightforward.
For $2,500 a year per camera, Flock would install, maintain, and operate the hardware. The city got access to a software platform with real-time alerts and searchable footage. So, no upfront capital expenditure, no specialist IT staff required. It was surveillance as a service. And for a small city with a limited budget, it looked like an extraordinary deal. And in the same way that the first month of any subscription is an extraordinary deal, the complications arrive later. And the venture capital market agreed. Flock has raised over $1 billion in total funding.
Its backers include Andreessen Horowitz, Founders Fund, Tiger Global, Kleiner Perkins, and Y Combinator. So, the usual people who show up when something is either going to change the world or get investigated by Congress, often both. As of April 2026, the company was valued at $8.4 billion.
Its annual recurring revenue crossed $300 million in 2025, representing 70% year-over-year growth. More than 5,000 law enforcement agencies and 4,800 police departments are connected to its network. That network is the critical architecture detail. Flock doesn't just sell cameras to individual cities, it connects them into a national system where any participating agency can search any other agency's camera data. A police department in Georgia can run a plate against cameras at a school in Texas. A sheriff's office in Indiana can search parking lot footage from a suburb in Illinois. No warrant needed, no subpoena, no notification to the city that owns the camera, no particular reason needed, frankly. The network effect is precisely what makes Flock attractive to investors. It is also precisely what should make everyone else nervous.
And there is a historical parallel here that I think is worth sitting with for a second. In the early 2010s, Facebook data sharing defaults kept quietly expanding. Third-party apps could access not just your data, but your friends' data. The platform setting were technically configurable, but the defaults favored maximum sharing because maximum sharing was maximum growth. Most users had no idea how broadly their information was being distributed until Cambridge Analytica made it impossible to ignore. By then, the data was already out, which is the tech industry's version of we've already eaten the cake, but we'd love your thoughts on the recipe, please.
The architecture had been doing its work for years before anyone with the authority to stop it understood what was actually happening. Flock's national camera network operates on a structurally identical logic to this.
The default is sharing, the value is in the network, and the people whose data flows through it have very little visibility into where it goes. If you've been following this channel, this pattern might feel familiar. I recently covered surveillance pricing in a a video essay that I will link below.
That's on the practice of using AI to charge different customers different prices based on their own personal data.
That one was deployment of AI-powered monitoring going sideways. This is another case. The connecting thread between these stories is not the specific technology, it's the deployment itself. AI is powerful and in many cases it is generally useful, but the incentive to deploy it in ways that expand surveillance keeps winning because expanding surveillance is where the money is. And when the financial incentive and the surveillance capability point in the same direction, the question is not whether somebody will abuse it, the question is just how quickly.
So, what went wrong? Well, three things and each one reveals a different dimension of the same structural problem. The first is immigration enforcement. Flock's own website states clearly, "Flock does not work with US Immigration and Customs Enforcement ICE." The company does not partner with ICE. ICE does not have direct access to Flock camera systems or data. That is technically true. Flock did not hand data to ICE. What Flock built was a system where 4,800 police departments can search each other's cameras with functionally zero oversight and some of those departments have agreements that turn local officers into immigration agents. The result documented by 404 Media, the University of Washington Center for Human Rights, and The 74 is that local law enforcement officers have been running Flock searches on behalf of ICE, so-called side door access. Over 4,000 lookups have been conducted at the behest of the federal government for immigration purposes. School cameras installed in over 100 public school systems are part of the searchable network. Dayton Ohio was sharing Flock data for immigration enforcement apparently by accident, which is the phrase that should never appear in the same sentence as immigration enforcement, and yet here we are. Flock does not work with ICE is doing a lot of work there. Flock built a highway. It is not responsible who drives on it. That distinction is, I suspect, more comforting to Flock's legal team than it is to the communities being surveilled.
The second failure is the Dunwoody incident. Dunwoody is a suburb of Atlanta where Flock operates a comprehensive surveillance system, including a real-time crime center powered by Flock Safety. A local resident named Jason Huynh filed a public records request and obtained the access logs. What he found was that Flock employees had been logging into Dunwoody's camera system to run live sales demonstrations for police departments in other cities. The cameras they chose to demo included feeds inside the Marcus Jewish Community Center of Atlanta, specifically cameras pointed at the children's gymnastics room, the pool, and the playground. A Flock vice president accessed the gymnastics camera. Another employee accessed cameras labeled Jim Mandel 1 and Main Pool Right. Flock's website states that nobody from Flock Safety is accessing or monitoring your footage. The CEO personally apologized to the community center and promised radical transparency. That is a bit like a locksmith who got caught letting himself into your house promising to be more transparent about his key collection.
Residents packed city council meetings for nearly 3 hours of public protest.
The city renewed the contract anyway.
So, 3 hours of public fury followed by the council presumably just shrug and basically saying, "But the price was really good." The third failure is in the contract trap itself. In February 2026, Flock rewrote its terms and conditions. The new terms had granted the company a perpetual irrevocable license to use and disclose all customer data. The previous version had explicitly stated the following, "Flock does not own and shall not sell customer data." That promise was deleted. Flock called the removal redundant in the same way that removing the locks from your front door could be called redundant if you already trust everyone on your street. As one resident who read the full contract noted, no replacement language exists anywhere. And meanwhile, Flock's marketing website still promises that your neighborhood owns 100% of the data, and Flock Safety will never share, sell, or access your data, which means either the marketing team and the legal team are not speaking, or they are and this is just the result.
Cities that want out are discovering that leaving is not exactly simple.
Contract terms are unclear on whether cities can unilaterally deactivate the cameras. In Evanston, the city terminated its contract and Flock reinstalled the cameras without the city's permission, leading to a cease and desist. In Eugene, Oregon, the city terminated its contract and requested Flock remove the cameras by December 12th. Flock told them it would not begin removal until January 26th. The city sent its own staff out to physically take the cameras down. The city literally sent casual workers out with ladders. This is literally what happens when your surveillance vendor ghosts your breakup texts. More than 40 cities have now suspended or terminated Flock contracts, including Mountain View, Cambridge, Santa Cruz, and Boston. Many more are debating it. And through all of this, the I saw the access, the children's gymnastics footage, the contract rewrites, the reinstated cameras, the 40 cities pulling out, Flock's valuation went from $3.5 billion to $8.4 billion. The market is rewarding the behavior that communities are protesting. That is not a bug. That is the incentive structure working exactly as designed.
By the way, if you'd like to support this work and help keep it free and without any gatekeeping, you can do that via coffee or channel memberships. The links are down below. And now I want to zoom out because Flock is a case study, but the pattern is bigger than just one company, and it connects to something I think most people fundamentally underestimate how much of their public life is already being recorded, identified, and stored. When stories like this break on this channel, the comment section usually fills up with a certain kind of response. I'll just pay in cash. I'll put my phone in a Faraday cage. I'll just use a VPN. And I understand the impulse, and it's not necessarily incorrect, but it reflects a model of surveillance that is about a decade out of date. The idea that if you control your device, you control your visibility. Flock's cameras don't need your phone. They don't need your name.
They build what the company calls a vehicle fingerprint, not just the license plate, but the make, model, color, bumper stickers, bike rack, dents, scratches, stuff like that. Pay cash at the shop if you like. The car you drove there has already been logged, matched, and timestamped across every Flock camera between your home and the parking lot. You were identified before you opened your wallet. And vehicles are just only one layer. Think about how many identification systems you pass through in a single day without noticing. You walk into a shop, the CCTV above the entrance is no longer just recording grainy footage to a VHS tape in a back room. Increasingly, it is connected to AI-powered platforms that can match faces, track movement patterns, and flag individuals in real time. You tap your card or your phone at the register, that transaction is logged with your name, the time, the location, and what you bought. You walk past a building with a smart doorbell camera, that footage is searchable. You get on public transport, more cameras. You drive through a junction, Flock or one of its competitors captures your vehicle fingerprint. You enter your workplace, access badge logged. You were identifiable at every single one of these points, and you didn't opt into any of them. Facial recognition databases already exist within government systems. Your passport photo, your driving license photo, your visa application, that biometric data sits in databases that law enforcement agencies can and do access, and it does not stop at government. Clearview AI, a separate company, has scraped over 60 billion images from the public internet to build a facial recognition database that law enforcement agencies pay to access. US Customs and Border Protection signed a contract with them. The layer [clears throat] stack as well, license plate readers, facial recognition, transaction records, phone location data, building access logs, smart home devices, each one is just a partial view, but combined, they are a comprehensive record of where you were, when, and often why. You would need to avoid every camera in every public space you enter, which in a modern city is functionally impossible. The only genuine dead zones are deep wilderness and the open ocean, and even there, if you brought your phone, you're identifiable anyway. Now, I'm not saying this to fearmonger, by the way, most of real life is still very normal. Nobody is being hunted through the streets by an algorithm on a Tuesday afternoon, but the infrastructure exists, the data exists, and the point of analyzing these things properly, which is what this channel is about, is to think through what happens when the incentives align for that infrastructure to be used in ways nobody's stress tested. Because we have already seen that happen. Flock's Network was built for stolen cars and missing persons. It is now being used for immigration enforcement, sales demonstrations involving children, and data sharing under terms that were rewritten after cities signed up. The technology itself, AI, is neutral, like fire or electricity. It can be used for extraordinary good, but deployment without guardrails means the use case that emerges is the one that makes the most money or serves the most power, not the one that serves the public interest necessarily. And in a competitive economic environment where someone has to choose between making the world slightly worse or making money, some people will just choose the money. I'll avoid the C-word here, not because I'm afraid of it, but because it tends to make the comment section lose its collective mind, and I need you guys focused for this one. But, increasingly, a lot of people, when pressed, will simply choose to make the world slightly worse if it means they can survive. And as economic pressure tightens further, more people get pushed towards that choice. It only takes one or two. One employee pulling up a gymnastics room feed for a sales demo, one local officer running a plate search on behalf of a federal agency, the system doesn't need to fail catastrophically, it just needs to fail in the small, predictable ways that nobody bothered to prevent. This brings me to regulation, and this is a point I think about a lot. Have you ever looked at a law and just thought, "Why does this exist? Who would ever do that?" In Alabama, there is a law explicitly banning bear wrestling, not because Alabama legislators woke up one morning worried about theoretical bear combat, but because people were actually training bears to fight and surgically altering them to be better wrestlers. In Illinois, it is a felony to possess more than $300 worth of salamanders, which raises the immediate question of how anyone arrives at exactly $300 worth of salamanders and what they were planning to do with them. In Mobile, Alabama, it is illegal to possess prey string, and you can picture exactly the kind of festival cleanup disaster that prompted that one. Every one of these laws sounds absurd until you realize the law exists because somebody already did the thing.
That is how regulation works. It is inherently reactive. Somebody does something nobody anticipated, and then the rule gets written. Right now, there is no law that prevents a surveillance company from using live footage of children's gymnastics classes as a sales demonstration. There is no federal regulation governing how data from a national network of 90,000 AI-powered cameras gets shared between agencies.
There is no requirement that cities be able to unilaterally deactivate surveillance hardware they are paying for. These rules do not exist because until very recently, nobody imagined they would need to. The The problem is that AI deployment moves at venture capital speed and regulation moves at committee speed. By the time the bin bag goes over the camera, the data has already been shared with 4,800 agencies, the contract terms have already been rewritten, the valuation has already doubled. The reactive model of regulation, wait for the damage then legislate, cannot keep pace with technology that scales this fast.
Privacy and computer science is a massive field. The researchers I met during my PhD had methods so sophisticated that even I, as somebody with a doctorate in the discipline, found them very non-trivial. The expertise exists, the methods exist, differential privacy, federated learning, access control architectures that would make side or search a structurally impossible. The technical solutions to these problems are not hypothetical. They are just not being required. I'm not saying that we can write perfect regulation, by the way.
The field of law is complex because humans are complex, but the current approach, deploy first, discover the misuse later, spend months in city council meetings debating whether you're allowed to remove the hardware, this is not good enough. What is needed is something closer to how software itself gets built. Iterative, adaptive, tested against failure modes before deployment, not after. Think of it as agile regulation. It sounds like a contradiction, I know, but it is the only model that has any hope of keeping pace because right now the bin bag is the most ominous symbol I've seen in technology governance in years. It is the government that bought a surveillance system it didn't fully understand, discovered it was being used in ways nobody anticipated, and responded with the only tool it had available. Not a policy, not an injunction, not a technical safeguard, a bin bag placed carefully over a camera that may or may not still be recording, installed by a company that may or may not honor a termination notice, feeding data into a network that no single city controls. That is what happens when the speed of deployment outpaces the speed of oversight. You don't get solutions, you don't get regulation, you get a bin bag and a hope that nobody checks whether the camera is still on or not.
But surveillance cameras tracking where you drive are only one layer of how AI is being used to watch you. What happens when the same logic gets applied to what you buy and companies start charging you different prices based on your personal data? The results are about as reassuring as a bin bag over a camera.
That's the video that I would watch next. Thanks so much for watching this one. Subscribe and I'll see you all on the next one.
Related Videos
BREAKING: Judge Kathleen Issues Emergency Arrest Warrant After Trump Defies Order
Frontora
2K views•2026-05-29
8 Hidden Things About Mackenzie Shirilla Netflix's 'The Crash' Didn't Show You
MarvelousVideos
2K views•2026-05-28
MP Garnett Genuis warns Canada’s MAiD system has ‘gone too far’
WesternStandard
187 views•2026-05-28
THE STREISAND EFFECT AT BARBARA STREISAND’S HOUSE! - First Amendment Audit
KULTNEWS
1K views•2026-05-30
Trump Impeachment STORM IGNITES as 29 Judges Vote for Conviction!!
DanielBriefDaily
2K views•2026-06-02
EBK Jaaybo Won’t Be Going To Trial?! | Criminal Lawyer Reacts
floridadefenseteam
404 views•2026-05-29
OFFICE HOURS: The Theft of Black Brilliance... AI and Intellectual Property (w/ Lisa E. Davis)
marclamonthillnetwork
2K views•2026-05-29
सुप्रीम कोर्ट में 5 जजों का शपथग्रहण समारोह #supremecourt #judges #oathceremony #shorts #ytshorts
Bharat24Liv
4K views•2026-06-02











