What the ‘Flock’!

A few weeks ago I took a joyride in Northern Virginia. I like driving fast, so my driving behavior during these joyrides usually test the bounds of legality. As such, my style of driving demands a proper awareness of speeding cameras and I’ve become very experienced with identifying them from afar. During this drive, however, I noticed something peculiar. This something was definetely a camera, but it didn’t resemble any speed camera I had ever seen. Nevertheless, I slowed down as I passed it, just in case, but I didn’t really think about it after that.

A few days ago I was browsing HackerNews and saw a post on Flock Cameras. Unaware of what this was, I clicked on the post and immediately recognized that camera from a few weeks ago. This was my first introduction to Flock Cameras, more generally known as Automated License Plate Readers (ALPRs), and after learning more about them, I have some thoughts.

For those who may not be familiar with them, automated license plate readers (ALPRs) are devices marketed as passive surveillance tools, primarily designed to identify and record the license plates of passing vehicles. While several vendors design and sell ALPR systems, Flock is by far the largest and most widely recognized provider accounting for roughly 83% of the ~119,000 ALPRs mapped across the U.S. (Fox News). This is why ALPRs are often colloquially referred to as “Flock cameras.” The customers who deploy ALPRs vary widely, ranging from local law enforcement agencies to large retailers such as Home Depot (The American Prospect) and even homeowners’ associations (Law and Society Review).

The technology behind ALPRs is relatively straightforward. At a high level, an ALPR can be thought of as a smart camera: it captures images or video of passing vehicles and uploads the recordings to cloud servers for processing. AI models then analyze these recordings and extract information, which is stored in a searchable database. This information includes not only the vehicle’s license plate, but also characteristics such as its make, model, color, and even distinguishing features like bumper stickers. A key technology enabling this capability is the convolutional neural network (CNN). CNNs are particularly effective at recognizing visual patterns. These networks can learn to detect simple features such as edges and lines in the early layers of the network and progressively combine those features into more complex objects, such as characters, license plates, and other vehicle characteristics (CNNs).

Flock describes their technology as one that can identify “unique vehicle characteristics”(Flock). This claim is interesting becomes the company seems to be boasting about their fingerprinting capabilities – their ability to identify and track an individual’s mobility patterns over time. This information can reveal a great deal about someone. If you think about the places you visit, the frequencies, the times, .etc, you could imagine that someone who had a searchable database of that information might know you as well as your close friends do. Maybe they could be reasonably confident about your profession, your age, income, hobbies, or other aspects of your life.

Ostensibly, ALPRs are only being used to solve crimes and keep neighborhoods safe. This is what proponents of these systems tell us. But, digging deeper, this may not fully be the case. In 2025, the Johnson County Sheriff’s office used Flock cameras to search for a women who was accused of having an illegal self-induced abortion. Over 83,000 cameras were used in this search, and these cameras belonged to entities across several states – states where abortions were legal. This search was based off suspicion…not because a crime was committed (EFF; Snopes).

Flock cameras have been used by ICE to identify individuals for detainment and deportation. Local police departments have conducted searches of the Flock databases on behalf of ICE agents who had no access to such systems. In Illinois, login credentials of police authorities were simply handed over to ICE agents so they could access the system (Reason). These are not isolated incidents either. Florida has recently revoked the use of Flock cameras on state highways due to similar privacy considerations and grievances (NBC News).

The concerning problem I see here is not necessarily the technology itself – it is a people problem, and ultimately a policy problem. There is a lack of accountability around who can access this sensitive information and what they can do with it. Movement data can reveal deeply personal details about an individual’s life, yet it is unclear what prevents an arbitrary person with access to these systems from piecing that information together – or from providing that capability to someone else. So, what can be done about this? My first thought is to throw more technology at the problem, specifically cryptographic tools. Secure multi-party computation (MPC) comes to mind, partly because it is the area I work in. MPC is a cryptographic technique that allows multiple parties to jointly compute a function and learn the output without learning the individual inputs of other parties. More relevant to the ALPR setting, variations of MPC exist that require all participating parties to contribute to the computation before the output can be learned.

In the ALPR context, for example, one party could be a local police department and another could be the local judicial office. Additional parties could be included as necessary. A query against the ALPR database would then require the participating parties to jointly authorize and execute it. This provides a form of accountability; if an unauthorized query is made, there is a clear record of which parties participated in approving it. It also provides a form of access control; neither party can simply bypass the other and run the query independently. If needed, the query can also be kept secret from other parties who shouldn’t have access to it.

This approach, however, may be overkill. MPC can introduce additional computational and operational costs, and simpler cryptographic mechanisms may provide sufficient protection. More importantly, without laws or policies requiring these protections, it is difficult to imagine police departments voluntarily redesigning their existing systems around MPC.

I bring up MPC not because I think it is necessarily the solution, but to make a broader point: technology for privacy-preserving surveillance already exists. The difficult part is recognizing the problems that a technology can create and deciding, before deploying it at scale, what protections should be in place to address those problems. We need an open discussion about how much privacy we are willing to give up, who should have access to sensitive surveillance data, and what accountability should look like. The technology should follow from those decisions and not the other way around.