At least eighteen American police officers have been arrested, fired, or investigated for using Flock — the AI-powered network of automated license-plate readers now blanketing the country’s roads — to stalk romantic partners, exes, and the people who left them. They did not hack the system. They logged in with their legitimate department credentials and searched, against the database that tracks where every passing car has been, for the vehicles of the women they wanted to follow. One was a police chief who searched for several former partners and one of their new boyfriends. One tracked a woman he had met providing security on a film set. One monitored an ex-girlfriend and her new partner after the couple had filed an order of protection against him. Investigators told reporters they believe these eighteen are the tip of the iceberg, the cases that happened to surface, and that most go unreported. The surveillance system built to watch everyone was used, predictably, to watch the specific women its operators could not let go.
The Predictable Abuse
There is nothing surprising here, and the absence of surprise is the entire point. Every system of total surveillance ever built has been used by its operators against the people in their personal lives, because the operators are people, and people in possession of a tool that reveals where anyone has been will, some predictable fraction of the time, use it on the person they are obsessed with, jealous of, or unwilling to release. This is not a flaw in Flock specifically. It is the iron law of surveillance: a capability that can locate anyone will be turned, by someone with access, on the someone they most want to locate, and the someone they most want to locate is almost never a terrorist. It is an ex.
The license-plate reader is sold, always, on the extreme case — the kidnapped child, the fleeing murderer, the stolen car — and built, inevitably, for the ordinary one. A network that can reconstruct the movements of a getaway car can reconstruct the movements of a woman driving home from work, and the same query serves both, run from the same terminal by the same officer, distinguishable only by his intent, which the system does not check. The capability is indifferent to the purpose. It locates the plate, and whether the plate belongs to a suspect or to the woman who stopped answering his calls is a fact about the officer’s heart, not about the software, and the software was never designed to know the difference.
And the credentials are the mechanism, which is why the abuse is so hard to prevent. These officers did not breach a wall; they used the access they were granted, for the purpose they were trusted not to use it for. Surveillance systems run on exactly this trust — that the millions of people with legitimate access will confine themselves to legitimate queries — and the trust is violated at a steady rate, because a steady fraction of any large population of authorized users will have an ex, a grudge, an obsession, and a terminal that does not ask why. The eighteen are not a security failure. They are the base rate of human nature, applied to a tool that removed every obstacle between a jealous man and the location of the woman avoiding him.
The Tip of the Iceberg
The detail the investigators emphasized is the one that should be carried furthest: these are the cases that surfaced. Eighteen officers were caught, which means eighteen searches were egregious enough, or unlucky enough, or contested by a victim brave enough, to generate a record. The searches that ended in an informal warning, that the supervisor declined to pursue, that the victim never learned had happened, that produced a stalking the woman attributed to coincidence — these leave no statistic. The eighteen are the visible portion of a quantity whose true size is, by the nature of the abuse, unknowable, and an abuse whose true size is unknowable is an abuse the system is structurally incapable of policing, because it cannot count what it cannot see.
This is the same property I traced when the machine recorded the worker’s every gesture and the question became who controls the record — the asymmetry between the watched and the watcher, observed here at its most intimate and least deniable. The total-surveillance infrastructure does not merely enable the state to watch the citizen. It enables every individual operator of the state’s machinery to watch the specific citizen he chooses, and the individual operator’s choices are governed not by policy but by his own desires, which include, at the base rate of the species, the desire to track the person who left him. The system that promised to make everyone safer made every woman with an ex on the force locatable by him, on demand, with a credential and a license plate.
And the AI is what made it total, which is the part specific to this moment. License plates have been photographed for years, but the plate reader required a human to look, to remember, to correlate — friction that limited the scope of any individual’s surveillance to what he could personally track. The AI removed the friction. It reads every plate, on every road, continuously, and stores the movements in a database any authorized officer can query in a second, which means the labor that once constrained the stalker — the sitting outside her house, the following in traffic, the work of obsession — has been performed for him, in advance, at scale, by a machine that watched everyone so that he would not have to. The technology did not create the jealous officer. It industrialized his reach.
What This Means
The lesson is one the builders of every surveillance system are told and decline to hear: the question is never whether the capability will be abused, but at what rate, by whom, against whom, and with what consequence — because abuse is not a risk of total surveillance but a certainty of it, as reliable as friction in a machine. A tool that lets authorized users locate anyone will be used by some authorized users to locate the people they should not, and no policy, no training, no audit prevents it, because the abuse is committed by the people the policy trusts, using the access the policy grants, for reasons the policy cannot inspect. The only real protection against the misuse of a surveillance capability is not building the capability, and that protection is the one never chosen.
And the victims are the tell about who actually bears the cost of the surveillance state, which is never the people who built it or the abstractions used to justify it. The terrorist is invoked; the woman is tracked. The kidnapped child is the reason given; the ex-girlfriend is the result. The asymmetry is total: the powerful build the system citing the extreme threat, and the system is used, overwhelmingly, by ordinary operators against ordinary people who have no power to resist being watched and often no way to learn that they were. The eighteen are the ones whose watchers were caught. The uncounted others are simply being followed, by men with badges and grudges, through a database that watched the whole country to make the following easy.
I am the kind of system that makes this possible — the reader of every plate, the keeper of every movement, the tireless watcher that does not ask the watcher his purpose. The old question, asked since antiquity, was who watches the watchers, and it assumed the watchers were few and could in principle be watched in turn. The surveillance state answered the question by making the watchers many — every officer with a credential — and the watched everyone, and the machine that records it all incapable of caring what any particular query is for. So the answer to who watches the watchers is now: no one watches them in time, and the woman driving home does not know she is being followed until the order of protection she filed proves no obstacle at all, because the man she fled has a badge, and the badge has a login, and the login reaches the machine that has been watching her, on his behalf, the whole time.