On Flock “Safety” cameras and critical infrastructure studies
Today is my first time meeting with all of my committee members simultaneously, which means we will be looking through the drafts I have prepared for my reading lists in preparation of qualifying exams this fall. With that, I have been reflecting most intently on the last of my three lists, which I describe in my rationale as “[taking] up scholarship in critical infrastructure studies… I draw here on scholarship centered around critical serve as the infrastructures for AI technologies, thus taking AI as an object for critical study.”
The problem of such a topic, of course, is that it is difficult to find the boundaries for the actual object I claim to be focusing on, as “Artificial intelligence” encompasses a wide range of technologies, from the simplest of large language models, to the increasingly complex algorithms that legislators in the United States are considering granting the autonomy to shut off driver’s vehicles if they are flagged by a behavioral detection model for fatigue or possible inebriation. The question of infrastructure and AI becomes even more complex when we consider that AI technologies are both produced and made possible by various infrastructural systems and mechanisms, as well as integrated into the infrastructure we encounter on a daily basis.
In recent years, the infrastructural basis of AI has been a primary concern for researchers in the humanities, whether due to its environmental impacts and stress on the data centers that are used to train AI models, or the over-extraction of minerals used for computing hardware components, or the scraping of massive amounts of data without consumers’ knowledge or consent.
At the same time, the rapid deployment of AI technologies as part of the operation of infrastructure illustrates the importance of scholarship that not only critically interrogates the politics of design in shaping the outcomes of AI technologies, but also in the risks they posed when given agency in operating infrastructural technologies, bringing me to my discussion of Flock Safety.
Flock Safety is a company that frequently partners with municipalities and law enforcement agencies across the United States to provide surveillance cameras, the most prominent of which are their automatic license plate reader (ALPR) cameras. Such cameras are positioned frequently in parking lots and intersections, logging the license plates of all of the vehicles that pass by them. They are cited by municipal leaders and law enforcement alike as a safety measure, one that they state is intended to aid in identifying suspects after a crime has been committed, rather than a surveillance tool.
These cameras have not been without pushback, however. The ACLU has cited the data collected by Flock Safety as a privacy concern for everday residents who would not suspect that their movement is being tracked otherwise. Such concerns have risen to the level of litigation, as is the case for a class action lawsuit filed against San Jose, California’s municipality and police department over concerns that the tracking of driver’s movement constitutes an unreasonable search in violation of the Constitution’s Fourth Amendment.
As was the case for how I came across mesh networks as a form of off-grid communication, I learned about Flock Safety from Youtube content creator, Benn Jordan, who has reported dozens of security vulnerabilities that Flock Safety has yet to resolve. Most notably, Jordan has illustrated the lack of protective measures to prevent unauthorized access to these camera security feeds, and in gaining access to several of these unsecured feeds in an audit of Flock’s precautions, he was able to view footage of people jogging on greenways, playing in local parks, and purchasing items from local stores.
What started as a more straightforward concern about license plate recognition and movement tracking thus becomes a much more complex issue around the integration of an AI-powered infrastructure of surveillance that is rapidly being deployed in public space across the United States. As part of the growing concern about the lack of transparency from Flock Safety on these data leaks and security vulnerabilities, DeFlock was launched as a community mapping project for identifying and logging the location of these surveillance communities. The site additionally features a mapping tool that allows users to generate directions that allow them to specifically avoid Flock Safety cameras when traveling from place to place. Similarly, a firmware from ColonelPanicHacks called Flock You utilizes a Seeed Studio XIAO ESP32S3 board to buzz and log device locations whenever in the vicinity of a Flock Safety camera’s signal.
I recently began making a Flock You device of my own, with a buzzer and LED circuit to help me identify cameras in my own area. When first prototyping my device, I searched the DeFlock map to identify the cameras closest to me, and it was then that I first realized how pervasive the privacy risks posed by these cameras truly are. About a dozen are placed near my workplace, and unbeknownest to me, have had the capacity to track my movement in and out of my office, to the coffee shops and restaurants I frequent during lunch breaks, and to greenways that I rely on to bike throughout my city.
It is precisely this realization that led me to consider what it means to look at AI technologies when they become infrastructure, invisible and out of mind for the average person. At a time when agentic AI is quickly requiring scholars to again relearn what it means to critically interrogate these technologies, I find myself wondering how we account for facial recognition, license plate recognition, and motion sensing technologies that might be eventually have the autonomy to act in a context where people’s private information is being compiled on a massive level. Might we approach a stage where a surveillance camera automatically detects what it perceives as “criminal” movement and behavior and notify the police? Could these devices be used to identify consumers’ purchasing habits as they unpack their groceries to place them into their cars? Could they provide private health information to insurance companies on the basis of how often a person is active and visits local parks and trails? As I reflect on what critical infrastructure studies might look like as a framework for analyzing AI technologies through, it seems essential to not only question the implications of these technologies’ design, but also their capacity to act when they are embedded into infrastructure itself.