Flock vs. Freedom: Has Modern Tech Rendered Our Privacy Laws Obsolete?

In a striking recent story, it was revealed in public records that a Texas police officer used the Flock camera network to hunt down a woman suspected of having a self-administered abortion. That this occurred in the current political climate of state-by-state curtailment of women’s access to abortion, and that the Flock query was accompanied by the somewhat dehumanizing search explanation of “had an abortion, search for female” played a role in this story catching public attention.

However, the system of software and hardware that underpinned the digital search for this “female” (the department alleged that the search was conducted because her family was concerned for her welfare, and they weren’t planning to prosecute her for the abortion) has been a matter controversy for years, dating to before the AI Era. As far back as 2019, the company Flock was defending their (then much smaller) network, arguing for its value to the police departments who employ it. Even at that time, Flock’s Closed Circuit Television (CCTV) network was stirring up debate and igniting fears of state power that could easily be abused.

At its core, Flock is a ALPR company, a maker and operator of Automated License Plate Readers. ALPRs have been around for several decades, and the older versions are rather less controversial, especially at automated toll stations, where they can help expedite payment. These ALPRs could be (and were) used by law enforcement to track vehicle movements. But early networks were often citywide or regional rather than nationally integrated, and the actual tracking was limited to your plate number and maybe your inspection sticker.

Flock cameras today are far more than just ALPRs. For starters, they look at a lot more than just your license plate. The artificial intelligence driven cameras can scan your vehicle for dents, scratches, bumper stickers (and their varying degrees of wear and tear), in addition to demographic and physical traits of the driver. All of this information is used to compile a vehicle profile. Many Flock cameras are not on highways but instead in areas where people also walk or bike. And the system behind the cameras, employing facial recognition, can build a profile of bikers and pedestrians as well. They record video (continuously, though snippets are often what is made available to police), not just still photos, and some Flock tools even record audio.

The most powerful part of Flock is, of course, the network aspect. There are nearly 100,000 cameras mounted in nominally public places around the country, and they all feed into a single national database of photos and videos that Flock manages. Both live feeds and a vast archive of past recordings spanning across years (remember Flock was operational back in 2019) are gathered together here, curated and sifted by powerful modern AI systems. Through these systems, a name, a plate, a photo (of a car, of a person, of anything really) can be uploaded by a participating law enforcement department as part of a query.

The internal search engine will then scour this massive database for any probable “hits” of that person or vehicle, spanning the entire country, tens of thousands of locations, perhaps over half a decade or more. It is worth noting that some content on the Flock website suggests that data is not stored indefinitely. However, I should mention a parallel case: Even during the period when a court explicitly ordered OpenAI to save all ChatGPT chats for their lawsuit with the New York Times, the ChatGPT system would often suggest user-deleted chats are deleted in OpenAI archives after only 30 days.

I personally tested this last year by asking ChatGPT if deleted chats are saved permanently, and had to repeatedly remind ChatGPT of the existence of this public court order. Many users might have been led to believe data was not retained when it was. And many other companies have a history of misrepresenting aspects of their data usage and retention. The point is, we should probably not assume that a nationwide surveillance system whose utility and business use-case grows with the size of their video and photo archive is deliberately deleting most of their content.

Flock’s actual business model is similar to that of Netflix: Selling subscriptions. Only, Flock sells to police departments, and every time a new town agrees to participate (usually authorizing the installation of new Flock cameras within their own town as their police department signs up for a subscription) the net of watched locations grows, and therefore, so does the usefulness of the Flock system to interested law enforcement agencies. In this way, growth of the company itself fuels future growth by improving the main product. Even if the price were three times higher, today’s Flock network is a far better deal for a department than the 2019 Flock network, due to the installation of literally tens of thousands of cameras, and massive improvements in the AI search tool.

Now, there are many articles out there detailing the current-day backlash against Flock cameras and similar corporate surveillance products that operate on the streets and sidewalks. Rather than simply re-hash all of that, I want to take this basic understanding of Flock as a company, and of their surveillance system and look at it in the context of America’s privacy laws and case precedents. Flock is very clear that their systems are legal, and that there is no expectation of privacy in public parks, on the street, on the sidewalk, or at a mall or business district. And this appears to be correct.

Unlike tracking a person’s cell phone data, which generally requires the police to obtain a warrant, recording video in a public space is a perfectly legal (and widely accepted) behavior, both for individuals and corporations. CCTV cameras in some form have been common in at least parts of America for over 40 years, and abundant since at least the turn of the millennium. The UK (and especially London) for a time was famously the most “watched” place on Earth, based on the per capita density of security cameras. This arose not due to 9/11 but the earlier era of the Troubles. Many of these cameras went up as early as the first half of the 1980s (before being replaced with better ones as the tech improved).

The idea that public spaces are usually fair game for video recording and photography is essentially rooted in the fact that other humans can see you there anyway, so you should probably behave with some discretion and responsibility, neither saying nor doing anything that you wouldn’t want others to witness. This is not an unfair standard in theory, and it makes some sense. But we really do have to appreciate just how the radically the preservation and distribution of records of one’s public behavior changed over time.

Two centuries ago, all that a public “incident” could produce was witness statements from people who saw it. By the end of the 19th century, photographs of your misadventures could be taken with portable cameras. In the mid-20th century, film cameras were widely available, and so were silent film video cameras. But most people didn’t carry either about, save for tourists, professional photographers, and the odd enthusiast. If you went to Central Park, NYC, on a random day in 1945, and did a dance routine in a bear suit, chances were decent no one would get it on film, unless you stayed there for longer than a few minutes. Maybe a tourist would snap a few photos or get 30 seconds of silent film. And when she developed those, perhaps two dozen people back home would ever see it. Twenty years later, in 1965, she could mail a copy to the local TV news, and maybe they’d air it between soap operas, but even if they did, it’s one and done: Miss that on TV and they’re not showing it again.

By the 1980s, in the West we start to see the proliferation of VHS-C videotape cameras, which importantly can record audio, not just video, and we see the rise of CCTV. Even by 1995, most people walking past you in the street, in NYC or a small town, have no camera with them at the moment, and the CCTV cameras in place are very different from today. The key difference in CCTV cameras between today and thirty years ago is not just the number of cameras, although that is a factor. There were already tens of thousands of CCTV cameras in public and privately-owned public spaces (like a store or mall) in the UK alone by 1999, and millions by 2004. Usage friction is the main difference: Unless you were watching the feed live in a control room, it was a lot of work to “see” when something happened.

Getting footage of a past event in the 1990s and 2000s depended on firstly, hoping the footage had not been overwritten, and secondly, trawling through vast amounts of video, many hours multiplied by the number of cameras. A UK Parliamentary report from 2002 demonstrates this problem very clearly: “In the case of the [1999] Brixton Nail Bomber over 4,000 hours were spent viewing 1,097 CCTV tapes. Some police forces map cameras in their area but the location of many cameras is unknown [to police]. Also, in line with the [UK] Data Protection Act, CCTV images must not be kept longer than is necessary.” They had to have human officers review the footage manually. And even in the 2000s, much of the CCTV systems still depended on VHS tapes to store the data, a system that intrinsically resists efficient centralization.

So the CCTV network in the US and other Western countries as well was heavily balkanized, every shop, mall, and town center running its own system. In addition, it was a great deal of work to review past footage: It had to be done manually, by having humans watch it. And lastly, the footage was not kept for long. If it’s 2001, the UK government (or US government) doesn’t have 500 hours of footage of you visiting the mall or taking the train from 1998. The system was clunky, slow, manual, and limited in its data capacity.

All of these factors came together, especially amid the Troubles (in the UK) and after 9/11 (in the United States) to lead a large portion of politicians and the public to feel that this system of surveillance — one that is decentralized, largely reactionary, requires extensive human labor, and doesn’t retain data forever, nor transfer it to third parties — was a fair sacrifice for our public safety. The same applied to CCTV cameras in schools, which became more common in the US in the 2000s, after Columbine. Even when the systems switched to digital, the data was stored on local hard drives, and the camera system had little capability to assess anything on its own. And storage space for the video recordings remained quite limited.

Again, it’s not the number of cameras that constrained the system’s power: One report found “in ­1999 that in the U.K., on a busy day in an urban environment, one person’s image might have been recorded by more than ­300 cameras from over­ 30 different CCTV systems.” But even though the cameras were watching you, they weren’t really paying attention, and your recorded actions would not be preserved forever. Pick a person, a random person out of millions in London or NYC in 2005. Most probably, the CCTV record built up throughout their day, week, or month will never be used for anything, or if it is, at least they will be only a background character in the frame. No one, not even NYPD, has the resources to use CCTV to build a profile of every person who makes an appearance on the cameras, every time they do so, and update it on a daily basis forming a complex dataset of millions of unaware, law abiding civilians. At least, no one could do this in 2005.

The main constraints are gone now. Flock’s network is all Flock-controlled, so decentralization is erased, at least for them (it very much remains an issue for millions of privately owned cameras across the country, especially older units). The data can now be saved for years, even decades, for every single camera, thanks to the explosive growth of data centers and powers of cloud storage. The footage is higher resolution than it was twenty years ago, and much higher quality than the VHS footage of the 1980s and 1990s. Most of all, the superpowers of artificial intelligence change especially the post-facto usage of footage entirely.

The largest CCTV manpower constraint was the people who have to watch the footage and pay attention to every minute. Now, artificial intelligence programs can watch all cameras in the network live, scan the archive of footage, and use facial recognition to identify every time every person has appeared in this vast video archive (all logged with each camera’s location, along with date and time), later tying a name and address to each person with other facial recognition search databases. This is the core change: Surveillance footage was constantly recorded in the past, but nothing was usually done with it. In 2005, sifting through 10 hours of footage from a thousand cameras was already a pain. Make that 100 hours for 3,000 cameras, and it would be done only for the gravest of criminal cases. And even then, police are probably looking for one person (or a few). If you are in the background and don’t fit the suspect profile, then you will still be ignored, not cataloged on the side.

No longer. Now that AI can do the hard, tiring labor of watching all that footage, no one flies under the radar. Every person in every frame of every Flock camera’s video recordings can be tracked, traced, and a comprehensive file assembled on them, just in case police departments ever take an interest in that person years from now. This changes the use case. Now, CCTV can be used as a mega-net to catch behavior anywhere in the country, not just a city-sized radius around the scene of a crime. Furthermore, the ease of searching means that the system is used to hunt people down for far lesser crimes than in the past would have merited the searching of so many hours of footage from so many cameras.

So, for instance, tracking a woman who gets an abortion all the way from Texas to Illinois and beyond goes from absurdly inefficient to highly practical, as fast as typing and hitting “Enter.” Imagine trying to conduct the officer’s “had an abortion, search for female” query in 2005! It would be impossible. Maybe they could check cameras around a certain Planned Parenthood building, but an operation like this one couldn’t be done.

You’ve probably noticed already, but reducing the workload of a large scale CCTV search (many cameras, many areas, over a large time window) by orders of magnitude makes it more likely that you could be the target of a search. If the effort involved is minimal, then the threshold to activate the system is much lower. Someone could accuse you in an email to the local Police Department in Texas of helping a woman obtain an abortion (and that, unlike self-administered abortion, is a real crime). If that department has a Flock subscription, they can now instantly search nearly 100,000 cameras across years of recording time to see if video clips or still images of your face, license plate, or car exist in a time and place that would incriminate you. Ten years ago, that search would have been too much work for such a low level crime. Perhaps Texas police officers with the tech of ten years ago would still investigate you, but there’s a much better chance they’d fall short of evidence sufficient for arrest.

There need not be an actual suspicion of criminal conduct. Flock can easily be used to build noncriminal personal profiles of millions or tens of millions of people. Even if this isn’t being presently done, there no real functional barrier to achieving it, nor is there a clear legal prohibition. Again, this could not be done under the old system, whether in 1985 or 2005. But now, every person clearly seen in frame of every photo and video the cameras record is readily studied and assessed.

Obviously this massive shift in the impact of a centralized recording network like Flock can and should change how we see the notion of public spaces being fair game for recordings. Being recorded passively by thousands of “dumb” cameras in 1985 or 2005, as a figure in the background who is ignored unless the subject of a major investigation is one thing. A fair sacrifice for security, so footage can be reviewed later and individual suspected of serious crimes can be identified. But being actively recorded, your every move cataloged and examined by a fallible and unaccountable AI system, where your every outing to CVS is added to The Database, is positively Orwellian and sits in a completely different class of intrusiveness.

In time, these perfectly legal cameras will not only obviate the need for cell data warrants in most urban and suburban regions (the Flock AI can track you camera to camera — though cell data will always add to the picture of someone’s movements), they will eliminate the right not to be tracked. Previously, with great (and ever growing) effort, one could avoid using modern computers and especially smartphones and smart watches, so that a GPS trail of your location would not exist for law enforcement to find. Now, on bicycle, on foot, on scooter, or in a car, you are tracked whether you like it or not. Even if you dressed up in a mask like it’s 2020 (illegal in some places and circumstances now), a sufficiently dense Flock of cameras could literally track you as you step off your property, all the way to through your errands, and back home. If they don’t know who, they still know where. To me at least, this is qualitatively different from 1990s or 2000s CCTV surveillance.

This new national camera network is far easier to abuse than any past version of the surveillance net. The pre-AI system, wherein hundreds of thousands of private businesses and homes (along with local, state, and federal government) owned millions of cameras, the footage was not saved forever, and it had to be reviewed manually if at all, was fairly good for personal privacy. It arguably achieved a fair balance between safety and our right to private lives. That balance was achieved through the difficulty of using what was captured on the screen.

If an authoritarian regime in 2005 America wanted to use the surveillance camera network of the era to spy on people it considered dissidents, the workload would be substantial. This would especially be true if the goal was to spy on them every minute they were out of the house. So many disparate private and local CCTV networks, thousands of hours of footage to trawl through every day, and the labor of organizing sightings into a timeline. Tailing the dissidents the old fashioned way would be more efficient! But that too would require several agents per dissident, constraining the number of people who could be thus surveilled. To paraphrase George Orwell on the topic of 1984’s “telescreens,” the regime could watch some people all of the time, and all people some of the time, but they can’t watch everyone all the time (though you never knew when you were being watched).

However, with enough Flock cameras and the power of AI, the regime can indeed watch everyone all the time, if they are outside their homes. And because each Flock camera is identified in the system with its location, the AI can easily make a space-and-time chart of where a specific person went and when, simply by estimating their distance from each Flock camera (and in which cardinal direction) every time a camera spots them. And this can be done all day, every day. It’s a level of pervasive state spying that an earlier generation of authoritarians could only dream about.

Even in democracies, this power to spy gets abused. A number of law enforcement officers have been charged with offenses relating to abuse of Flock to track ex-girlfriends or other individuals for personal reasons. To do this with the CCTV network of 2005 would be difficult if not impossible. Now they can do it with a quick search, tracking not only someone’s vehicle but the person themselves, perhaps to track them to the house of their new partner, for instance. Potentially for every case of this that gets reported there are several that do not, either because no one ever realized that person had been searched on Flock, an excuse was provided for the query, or because someone who did find out they’d been “Flocked” many times and suspected wrongdoing was wary of reporting a police officer for the conduct. Whether this unsettling abuse of the system has led to violence is unclear, but it could in the future, if potentially violent and abusive people are not stopped from using it to constantly track ex-partners.

Now, I don’t have a clear or easy solution to all of this. One suggestion is that we could re-write privacy laws to require CCTV networks to operate along the lines of a 2000s system, just with better video resolution: Don’t store the footage forever, don’t use AI to scan millions of hours of video to build profiles of millions of innocent people, don’t build a file on someone unless they are already suspected of wrongdoing. But there’s no decentralizing the network. Flock will always own their cameras. And that network will grow by tens of thousands each year.

Of course, rolling back the powers of the CCTV net will inevitably allow some horrific violent crimes to go unpunished, and it will arguably enable others to happen: Recall that I observed the modern surveillance net allows for people to be tracked for far lesser crimes than they would have before? Well, some of those people, caught today for a lesser offense because of how easy it is to search and find with Flock, would be investigated once arrested, and in some cases charged with other crimes once connected to more offenses. Serving time after all those convictions, these criminals will lose the opportunity to commit some additional future violent crimes.

We would have to make sacrifices in order to meaningfully constrain the AI supercharged spying state. There will be kids who go missing or are taken who would be found by Flock’s 250,000 cameras circa 2030, but cannot be spotted by the primitive systems of the 2000s, even with millions of actual cameras. The capacity to quickly search that many cameras, for so many hours of footage is just so powerful, it puts Flock (and similar systems) in another class. The further back in the timeline of the last century you look, the easier it was for people to get away with heinous crimes. Part of that is forensics, but part of it is the cameras.

Where once a human had to think to “check the cameras,” where once the footage could be overwritten after a week, now there is a growing AI driven network that keeps it all forever, and can rapidly search hundreds of petabytes with the computing power of modern data centers. Our society must decide if the existence of this spying and tracking machine is worth it or not. A country where everyone is watched all the time is a hard place to get away with crimes. It’s also a bad place to be in if a dictator takes over and turns The Machine against his rivals and their supporters. Can we confidently say that’s a far-fetched scenario?