The Suspicion Engine: How Flock Cameras Decide Who Police Should Watch

Consider the moment before a police officer knocks on a door. For most of the history of Anglo-American policing, that moment had a shape the law could recognise. Something happened. A crime was reported, a complaint was made, an officer saw an act with their own eyes, a witness named a name. The suspicion ran from a particular event to a particular person, and somewhere in the chain stood a human being who could be asked, afterwards, why they had acted. The chain could be reconstructed. It could be questioned in a courtroom. It could be wrong, and when it was wrong there was, at least in principle, someone to hold to account.

Now consider a different moment. A camera mounted on a pole on a residential street photographs the rear of every car that passes it, reads the plate, records the time and the location, and uploads all of it to a national database. The car belongs to no suspect. No crime has been reported. The driver has done nothing but drive. And yet, somewhere in a data centre, an algorithm owned by a private company is comparing the pattern of that car's movements against the patterns of every other car in a network of more than a hundred thousand cameras, and deciding, on criteria it will not disclose, whether the pattern looks like crime. If the algorithm decides that it does, it generates an alert, and the alert goes to the police. There was no event. There was no complaint. There was no human in the chain who chose to look at this person. There was only the pattern, and the machine's private judgement about what the pattern meant.

This is not a thought experiment. It is a description of features that Flock Safety, an Atlanta-based surveillance company, had built into its network by the summer of 2025, and that the American Civil Liberties Union documented in a piece of analysis published on 23 July 2025 by Jay Stanley, a senior policy analyst with the organisation's Speech, Privacy, and Technology Project. The features have anodyne names: “Multi-State Insights,” “Linked Vehicles,” “Convoy Search.” What they do is not anodyne at all. They mark the point at which a surveillance system stopped being a tool that police could query about a specific vehicle they already suspected, and became a system that decides, on its own, which ordinary people the police should begin to suspect. Stanley put the inversion plainly. The system, he wrote, is now “actively evaluating each of us to make a decision about whether we should be reported to law enforcement.” The question this raises is not a narrow one about one company's product roadmap. It is a question about what has happened to the relationship between the state and the person going about their day, and about who, if anyone, is answerable when the machine that now stands at the start of that relationship gets it wrong.

The Network You Did Not Consent To

To understand the scale of what has been built, begin with the infrastructure, because the infrastructure is the precondition for everything else. Flock Safety is a private company founded in 2017. By the middle of 2025 it had, according to its own marketing and to the ACLU's reporting, deployed close to ninety thousand automatic licence plate reader cameras, organised into roughly seven thousand networks, serving thousands of law enforcement agencies across more than four thousand American communities. The number has not stood still since: by July 2026 the company was claiming more than 120,000 cameras. The cameras capture, by the company's own account, over a billion images each month. They do not photograph only the cars of people under investigation. They photograph every car. The system is, by design, a record of the movements of an entire driving population, assembled continuously, retained for a period set by policy rather than by suspicion, and made searchable across jurisdictional lines that the people being recorded never agreed to cross.

The ordinary defence of such a system is that a licence plate is displayed in public, that there is no reasonable expectation of privacy in a number plate visible to any passer-by, and that a camera merely automates what a police officer standing on a corner could lawfully observe. This argument has a long pedigree and a narrow truth. A single observation of a single plate is not, in itself, a search in the constitutional sense. But the argument collapses under aggregation, and aggregation is precisely what the network performs. The officer on the corner sees one car once. The network sees every car everywhere, all the time, and remembers. It can reconstruct, from its archive, where a given vehicle has been for weeks or months. It can place a person at a clinic, a mosque, a union hall, a lover's house, a protest, a lawyer's office. The difference between one observation and a permanent, searchable, cross-referenced chronicle of a life in motion is not a difference of degree that can be waved away. It is the difference the Supreme Court of the United States recognised, in a different but closely related context, when it decided Carpenter v. United States in June 2018.

What Carpenter Saw Coming

In Carpenter, the Court confronted the government's warrantless acquisition of 127 days of historical cell-site location information belonging to a robbery suspect, Timothy Carpenter, from his mobile network provider. Chief Justice John Roberts, writing for a five-to-four majority, held that obtaining this record was a search under the Fourth Amendment and required a warrant. The reasoning matters more than the result. The Court found that location data of this kind creates “a detailed chronicle of a person's physical presence compiled every day, every moment over several years,” and that the mere fact the data was held by a third party did not strip it of constitutional protection. The decision is widely read as the Court's tentative embrace of what scholars call the mosaic theory: the idea that surveillance must be assessed in the aggregate, because a sufficiently complete picture of a person's movements reveals things no single observation ever could, and reveals them to a degree the framers of the Fourth Amendment would have regarded as the work of a tyranny.

Flock's network is, in every respect that Carpenter found troubling, a more comprehensive instrument than the cell-site records at issue in that case. It is location tracking by another name, conducted at population scale, retained and made searchable, and assembled by a private actor who then sells access to the state. The law has been slow to catch up. Carpenter was carefully, even nervously, limited to its facts; the Court declined to disturb the older third-party doctrine wholesale, and lower courts spent the years after it arguing about how far its logic reaches. That argument has now moved. On 29 June 2026, in Chatrie v. United States, the Court held by six votes to three, with Justice Elena Kagan writing for the majority, that obtaining a person's location history through a geofence warrant is a Fourth Amendment search, and that people keep a reasonable expectation of privacy in their location history even when it is held by a third party such as Google. The Court vacated and remanded without deciding whether the particular warrant was valid, but the third-party doctrine came out of the case meaningfully narrower than it went in.

Within months a federal judge had carried that reasoning to Flock. In United States v. Kyle, decided in the Northern District of Oklahoma in early October 2026, Judge Sara E. Hill considered the case of a Tulsa County deputy who had run a warrantless Flock query on a car for no better reason than that it carried California plates. The query returned more than fifty sightings across several states over roughly a month. Relying on Chatrie, Hill held that the search, conducted without a warrant or probable cause, violated the Fourth Amendment; she called Flock “a type of indiscriminate mass surveillance” and suppressed everything obtained afterwards as “fruit of the poisonous tree”. She was candid that nearly every court to consider the question had held that using licence plate readers is not a search, and hers is a single district-court ruling, binding on no other court, with an appeal expected. In September, in the Central District of California, Judge Virginia A. Phillips had reached the opposite conclusion, holding that querying the Flock database is not a search at all. The law is moving, then, but it is split. And Kyle, for all its force, concerned the older kind of intrusion: an officer querying a database about a car he had already chosen to look at. The moral logic of the decision points in one direction only. A system that can tell the state where you have been, continuously, without a warrant and without a predicate, is the thing Carpenter was worried about. And the new AI features take that system somewhere neither Carpenter nor Kyle contemplated.

From Query Tool to Suspicion Engine

Here is the distinction on which everything turns. Until recently, a licence plate network, however vast, was a reactive instrument. A crime occurred; an investigator had a reason to look for a particular car; the investigator queried the database to find where that car had been. The suspicion came first, generated by a human being in the world, and the database served it. This is intrusive, and the intrusion is real, but it preserves the traditional order of operations. A person decides there is a reason to suspect, and the machine answers a question that the person's suspicion has framed.

The features the ACLU documented in July 2025 invert that order. “Multi-State Insights” scans the entire network for vehicles that appear across multiple states and flags them as potentially linked to trafficking or organised crime. “Linked Vehicles,” also marketed as “Convoy Search,” lets the system surface vehicles “frequently seen together,” mapping associations between people who may have done nothing but park near one another or travel the same road at the same time. The point is no longer to answer a question a human investigator has asked. The point is for the algorithm to decide, by reading the movement patterns of people against whom there is no complaint and no predicate, which of them ought to become objects of police attention. As Stanley wrote, the system has shifted “from allowing police to find out more about specific vehicles of interest, to using the system to generate suspicion in the first place.” The suspicion no longer originates in a human mind responding to an event. It originates in a proprietary model responding to a statistical shape.

This is the privatisation of probable cause, and it is worth being precise about why that phrase is not hyperbole. Reasonable suspicion and probable cause are, in the American constitutional tradition, the thresholds the state must cross before it may stop, search, or arrest. They are meant to be reasoned judgements, articulable in words, defensible after the fact, and grounded in particular facts about a particular person. The entire architecture of accountability in policing assumes that somewhere there is an officer who formed a suspicion and can be made to explain it. When a private algorithm performs the first act of suspicion, that architecture has nothing to grip. The officer who receives the alert did not form the suspicion; the company that built the model did, in code, months earlier, on training data the officer has never seen. The suspicion arrives pre-formed, wearing the borrowed authority of computation, and the human at the end of the chain becomes a ratifier of a judgement made elsewhere by a system answerable to no one.

The Accountability Vacuum

A police department, for all its faults, is a public body. It is subject, in most American jurisdictions, to open records laws that allow citizens and journalists to demand its documents. It answers, at one remove, to elected officials: mayors, city councils, sheriffs who must stand for re-election. It can be sued, audited, subjected to consent decrees, investigated by inspectors general. These mechanisms are imperfect and frequently evaded, but they exist, and they exist because a democratic society decided that the power to deprive people of liberty must be held by institutions that can be made to answer for how they use it.

Flock Safety is none of these things. It is a private company. Its algorithms are trade secrets. The logic by which “Multi-State Insights” decides that a travel pattern is suspicious, the data on which the model was trained, the rate at which it produces false positives, the demographic distribution of the people it flags: all of this is proprietary and undisclosed, and because the company is private, it is not subject to the open records laws that govern the police departments it serves. There is no Freedom of Information request that reaches the inside of the model. There is no elected official with authority over its design. The ACLU's Stanley put the structural problem at the centre of his analysis: because “Flock is a private company not subject to checks and balances such as open records laws,” the public knows essentially nothing about how the suspicion is manufactured. The state has outsourced the first and most consequential act of policing — the decision about whom to watch — to a vendor that operates outside every accountability mechanism the state itself is bound by.

The consequences of this arrangement are not theoretical, and they did not stay confined to the marketing brochures. Through 2025 and into 2026, a series of investigations and municipal audits revealed that the network's data was flowing in directions its local operators had never authorised. Audits found that federal immigration agencies were running searches against local Flock systems, often through “nationwide lookup” features that gave agencies in other states access to a city's data despite local settings nominally restricting it. The city of Dayton, Ohio, discovered through its own review that more than seven thousand searches of its Flock cameras had been conducted for immigration enforcement purposes that the city's own policy explicitly prohibited; by June 2026 city workers were covering the cameras with bin bags while officials worked out what to do. An audit of the Oxnard, California, system found that a vendor-enabled nationwide query had exposed the department's data to outside agencies despite security settings configured for California-only access. By the spring of 2026, according to reporting compiled across multiple outlets, at least eighty American cities had cancelled, paused, or declined to renew their Flock contracts, with municipalities including Austin, Denver, Evanston, and Santa Cruz among them. Over the summer the trickle became a rout. The Oakland privacy group Secure Justice, whose tally is its own rather than any official count, recorded more than ninety jurisdictions ending their relationship with Flock in August 2026 alone, and 214 since 2021. Eleven Arizona cities terminated their contracts. In Texas, after a Texas Tribune investigation found that the state's Motor Vehicle Crime Prevention Authority had put at least 30 million dollars into some 3,200 Flock cameras since 2023, Governor Greg Abbott ordered state agencies in late August to pause funding for them, and Dallas moved to cancel 321 grant-funded cameras with effect from 15 September. In October, Henrico County, Virginia, ended its contract early, its county manager citing privacy concerns. The company has felt it: in October Reuters reported that Flock planned to cut about 18 per cent of its workforce, roughly 270 of some 1,500 employees, following a voluntary buyout in September, a figure Flock has not confirmed. The lesson of these episodes is not merely that a particular policy was violated. It is that once the data exists in a private national network, the local democratic controls that supposedly govern it are a polite fiction. The data goes where the network architecture and the company's commercial decisions permit it to go, and the city that installed the cameras learns where afterwards, if at all.

When the Machine Is Wrong

Every system that generates suspicion will sometimes generate it falsely, and the question of what happens to the falsely suspected is the question on which the legitimacy of the whole enterprise rests. With Flock's network, the answer is already being written, case by case, in the bodies of people who did nothing.

In Sherwood, Arkansas, in early 2026, a Flock camera flagged a black Chevrolet Tahoe as stolen. An officer pulled the vehicle over in a Kohl's car park, drew his weapon, and detained the driver. A second officer arrived. A female passenger in the rear seat was detained as well; beside her was a one-month-old infant. The plate had been misread by a single letter. In Morristown, Tennessee, a couple were stopped at gunpoint with their three-year-old granddaughter in the car after a Flock camera misread their plate. In Toledo, Ohio, a man was mauled by a police dog and jailed after an erroneous read. And in San Diego, on 26 November 2025, a Flock reader captured an image of a red Alfa Romeo on Moore Street in Old Town. A detective, Gary Gonzales, identified it as the vehicle police had been pursuing in connection with an attempted carjacking, citing the red paint and tinted windows. The car belonged to Hugo Parra, who was five miles from the scene, at a cigar lounge he had driven to directly from his apartment with a companion, Ariel Beltran. Parra spent roughly a month in jail before the charges were dropped. In April 2026 he and Beltran filed tort claims seeking 1.5 million dollars each, alleging false arrest, civil rights violations, and negligence, and contending that police had misread their own surveillance system and ignored exculpatory evidence in a rush to judgement. The City of San Diego has since denied both claims, and reporting indicates that the two men are preparing to sue for negligence and civil rights violations.

These are misreads of the older, query-driven kind — a plate matched against a hot list, a human jumping to a conclusion the data did not support. They are bad enough, and they already demonstrate the central pathology: the machine's output is treated as authoritative, the human officer defers to it, and the person on the receiving end pays in handcuffs, jail time, and trauma for an error they had no way to anticipate or prevent. Now extend the logic to the new features. When the suspicion is generated not by a plate matched against a list but by an opaque model's judgement that a travel pattern looks like crime, the falsely suspected person has even less recourse, because there is no discrete error to point to. There is no misread letter. There is only a pattern, classified by a proprietary system, on criteria no one will reveal. How does a person prove they are not the shape the algorithm mistook them for, when they are not permitted to know what shape it was looking for?

This is the accountability gap in its purest form. When the algorithm is wrong, responsibility dissolves across a chain in which no link will accept it. The officer says he was acting on an alert from a trusted system. The department says it relies on the vendor's technology. The vendor says its product is a tool, that the decision to act always rests with a human officer, and that its algorithms are proprietary and cannot be disclosed. Each actor points to another. The person who was stopped, jailed, or terrorised is left holding an injury that the system is structured to ensure no one owns. The human-out-of-the-loop design does not merely remove a safeguard. It removes the locus of responsibility itself, and it does so by design, because diffused responsibility is, for every institution in the chain, a feature rather than a bug.

The Testbed

There is a further dimension to this, and it is the one that most clearly exposes whose interests the system serves and whose bodies bear its costs. New surveillance technologies are not tested everywhere at once. They are tested somewhere first, and the somewhere is rarely a wealthy, white, politically connected neighbourhood with the lawyers and the city-council access to object. In Atlanta, the reporting outlet Capital B News documented, in an investigation published in April 2026 by Adam Mahoney and Madeline Thigpen, how the city's predominantly Black neighbourhoods in South Atlanta had become a live testing environment for the next generation of AI policing tools.

The picture they assembled is stark. Atlanta, by their account, has roughly 124 surveillance cameras for every thousand residents, a density higher than any city outside China. At the centre of the expansion sits the Atlanta Public Safety Training Center, the eighty-five-acre facility built on what was the South River Forest and known to its many opponents as Cop City, which includes mock city blocks wired with cameras, licence plate readers, and real-time crime centre feeds — a physical proving ground for surveillance systems. Records requests revealed that the Atlanta Police Department had tested Flock add-ons allowing officers to search the network's images and video using plain-language descriptions. Shruti Lakshmanan, a policy advocate with the ACLU of Georgia, described one such capability as the ability to pull up, for instance, every car bearing a particular political message on its bumper, and warned that the technology could be turned to targeting people for their beliefs. “AI is almost supercharging what mass surveillance can do,” she told the outlet.

The geographic pattern is not incidental, and it is not unique to Atlanta. The Capital B reporting cited a ten-city Harvard University study finding that surveillance cameras cluster most densely in gentrifying, diversifying neighbourhoods, with installations increasing as white residents move in, even after controlling for crime and income. Read alongside the bias concern Stanley raised in the ACLU's July analysis — that a model trained on historical crime data, itself the product of decades of disproportionate enforcement in Black neighbourhoods, will learn to treat residents of over-policed communities as inherently suspicious — the racial-justice dimension comes into focus. A suspicion engine trained on the past will reproduce the past. The patterns it learns to flag are the patterns of where police have already been looking, and police have been looking, disproportionately and demonstrably, at Black communities. The algorithm does not introduce the bias. It launders it, converting a history of discriminatory enforcement into the neutral-seeming language of a movement pattern that “looks like” crime, and then presenting the result to officers with the unearned authority of mathematics. The community that has been most heavily policed becomes the community the machine learns to suspect, which generates more enforcement, which becomes more training data, which deepens the suspicion. The loop closes, and it closes around the people who had the least power to object to its construction.

The Presumption Inverted

Step back from the particular features and the particular harms and the deeper transformation becomes visible. The presumption of innocence is not merely a rule about who bears the burden of proof in a criminal trial. It is a statement about the default posture of the state toward the citizen. In a society that takes the presumption seriously, the ordinary person going about their day is presumed to be doing nothing wrong, and the state must develop a particular, articulable reason before it may treat any of them as a suspect. Suspicion is the exception, requiring justification. Innocence is the default, requiring none.

The suspicion engine inverts this. It does not wait for a reason to suspect any particular person; it subjects everyone to continuous algorithmic evaluation and waits for the pattern that will single someone out. Every driver is, in effect, enrolled in a permanent line-up, perpetually being assessed for whether their movements have crossed some undisclosed threshold of seeming-suspicious. The default posture is no longer that the citizen is presumed innocent and left alone. The default posture is that the citizen is continuously analysed, and the question is merely whether and when the analysis will produce an alert. Innocence has stopped being the resting state and become a probationary status, conditional on never matching the pattern, the contents of which one is not permitted to know.

What changes about the relationship between the state and the ordinary person, then, is the thing that the older, event-driven model of suspicion was built to protect: the citizen's right to be left alone unless and until they give the state a specific reason to attend to them. That right is not a technicality. It is close to the whole of what distinguishes a free society from a surveilled one. A free society polices acts; a surveillance society polices patterns, and a pattern is something you can fall into without doing anything wrong at all. When a private, commercially motivated algorithm performs the first act of suspicion, the citizen loses not only privacy but the more fundamental thing privacy was protecting: the presumption that they are nobody's suspect until they have actually done something to deserve the name.

What Accountability Would Require

The grassroots response has already begun, and its shape suggests what genuine accountability might look like. In Colorado through the spring of 2026, volunteers fanned out across the state to map Flock's expanding footprint, standing beside traffic poles in towns such as Firestone and Golden with hand-written signs, documenting cameras installed through partnerships with nearly a hundred police agencies and more than a hundred and sixty private entities. The Colorado Sun chronicled the campaign in late April 2026. A legislative effort accompanied it: Senate Bill 70, sponsored by State Senator Judy Amabile, a Boulder Democrat, sought to place what she called “basic, common-sense guardrails on license plate reader data and how it's used, and protect the privacy rights of Colorado citizens.” It failed. After an intense lobbying campaign by law enforcement and a veto threat from Governor Jared Polis over the bill's warrant provisions, the measure was laid over until July, effectively killing it for the session. The pattern is familiar to anyone who has watched surveillance regulation attempted: the technology spreads faster than the law that would govern it, and when the law finally arrives, the institutions that have come to depend on the technology mobilise to stop it.

In Atlanta, advocates have pushed for the city to become the first in the American South to adopt a Community Control Over Police Surveillance ordinance, requiring public hearings, impact reports, and a city council vote before any police agency acquires or expands surveillance technology. This is the right structural instinct. The decision to build a suspicion engine over a community is a political decision, with profound consequences for the people who live under it, and political decisions belong to the public and its elected representatives, not to a procurement officer signing a contract with a vendor.

The city has since moved, though not as far as those advocates want. In September 2026, after an internal Atlanta Police Department audit of 115,578 Flock searches flagged seventy-nine for review, Mayor Andre Dickens introduced a citywide surveillance-technology ordinance. It would codify monthly internal audits, quarterly public summaries, an annual independent review of high-risk systems, and a public inventory of the city's surveillance tools, and it would require vendors such as Flock to write the city's privacy, retention, and data-sharing limits into their contracts. As of late September it sat in committee. The ACLU of Georgia says it falls short of what a CCOPS ordinance would require, and has urged the council not to renew the Flock contract at all. Audits and inventories are better than nothing, but they are oversight after the fact; they do not give the public a vote on whether the suspicion engine should be built in the first place.

A serious framework would go further. It would require, as a condition of any public contract, that the algorithms generating suspicion be disclosed to independent auditors with the authority to demand the training data, measure the error rates, and publish the demographic distribution of who gets flagged. It would prohibit human-out-of-the-loop suspicion generation entirely, restoring the principle that a named, accountable human being must form and be able to articulate the suspicion that precedes any police action. It would extend open records obligations to any private company performing a core governmental function, on the principle that one cannot privatise one's way out of constitutional accountability. It would impose strict, short retention limits, so that the network ceases to be a permanent chronicle and becomes, at most, a momentary observation. And where these conditions cannot be met — where a system is so opaque, so prone to error, or so structurally biased that no amount of procedure can render it compatible with a free society — it would ban the system outright, as a growing number of cities have effectively begun to do by tearing out their cameras.

The Answer to the Question

Who is responsible when the algorithm is wrong? Under the arrangement that exists today, the honest answer is: no one, by design. The officer defers to the alert. The department defers to the vendor. The vendor invokes trade secrecy and the fiction of the human decision-maker it has carefully positioned at the end of the chain to absorb the liability it has engineered itself out of. The person who was wrongly flagged, wrongly stopped, wrongly jailed, is left to file a tort claim, watch the city deny it, as Hugo Parra has, and prepare to sue in the hope that a court will eventually assign a responsibility that the system was structured to ensure no one would ever have to accept.

That answer is intolerable, and its intolerability is the whole point. A society does not get to deploy the power to manufacture suspicion at population scale and then arrange its affairs so that no one owns the consequences when the power is misused. The suspicion engine has not merely added a new tool to policing. It has dissolved the chain of human responsibility that made policing answerable in the first place, and it has handed the first and most consequential act of that policing — the decision about whom to suspect — to a private company motivated by the sale of more cameras and more subscriptions, operating outside every mechanism a democracy has built to hold power to account. The thing that has changed about the relationship between the state and the ordinary person is that an unaccountable commercial algorithm now stands between them, deciding who among the innocent will be treated as a suspect, on criteria no one will reveal, with no one to answer when it is wrong.

The citizens standing beside traffic poles in Golden and Firestone with their hand-written signs understand what is at stake more clearly than the institutions that have rushed to deploy the technology over their heads. They are not Luddites objecting to cameras. Nor, whatever the company would prefer to believe, are they bots. On 26 October 2026, at the International Association of Chiefs of Police conference in Orlando, Flock's chief executive, Garrett Langley, told police chiefs, in audio obtained by 404 Media, that the opposition was largely “anonymous bots, and so it is not real”, claiming that social media chatter about the company was down about 95 per cent from July. He offered no evidence. The volunteers in Colorado stood in the open with their names and their signs; the cancellations in Henrico, Dallas, and across Arizona are matters of public record. They are people who have grasped that a society in which a private machine decides who comes to the attention of the police, and in which no one is responsible when it errs, is not a society that has merely upgraded its policing. It is a society that has quietly relocated the power to suspect from accountable public institutions to an unaccountable private one, and called the relocation a feature. The work of the moment is to relocate it back, to insist that suspicion remain a human judgement made by people who can be made to answer for it, and to refuse the comfortable fiction that an algorithm's verdict is anyone's fault but the people who chose to believe it.

References

Jay Stanley. “Surveillance Company Flock Now Using AI to Report Us to Police if it Thinks Our Movement Patterns Are 'Suspicious'.” American Civil Liberties Union, 23 July 2025. https://www.aclu.org/news/national-security/surveillance-company-flock-now-using-ai-to-report-us-to-police-if-it-thinks-our-movement-patterns-are-suspicious
ACLU of Massachusetts, Data for Justice Project. “Flock Gives Law Enforcement All Over the Country Access to Your Location.” 7 October 2025. https://data.aclum.org/2025/10/07/flock-gives-law-enforcement-all-over-the-country-access-to-your-location/
American Civil Liberties Union. “How to Pump the Brakes on Your Police Department's Use of Flock's Mass Surveillance License Plate Readers.” https://www.aclu.org/news/privacy-technology/how-to-pump-the-brakes-on-your-police-departments-use-of-flocks-mass-surveillance-license-plate-readers
Adam Mahoney and Madeline Thigpen. “Atlanta's Cop City Combines AI Policing And Environmental Racism.” Capital B News, April 2026. https://capitalbnews.org/atlanta-cop-city-ai-policing-black-community/
The Colorado Sun. “As AI license plate readers spread across Colorado, grassroots movements push back.” 27 April 2026. https://coloradosun.com/2026/04/27/ai-license-plate-readers-flock-colorado-grassroots-movements-push-back/
Colorado Newsline. “Colorado lawmakers kill bill to limit police use of Flock camera data.” 29 April 2026. https://coloradonewsline.com/2026/04/29/colorado-lawmakers-kill-bill-to-limit-police-use-of-flock-camera-data/
Boulder Reporting Lab. “Colorado bill to limit Flock and other license plate readers fails after law enforcement pushback.” 30 April 2026. https://boulderreportinglab.org/2026/04/30/amabile-pulls-colorado-bill-limiting-flock-and-other-license-plate-readers-we-didnt-have-the-votes/
Carpenter v. United States, 585 U.S. ___ (2018). Constitution Center case library. https://constitutioncenter.org/the-constitution/supreme-court-case-library/carpenter-v-united-states
NPR. “Why some cities are canceling Flock license plate reader contracts.” 17 February 2026. https://www.npr.org/2026/02/17/nx-s1-5612825/flock-contracts-canceled-immigration-survillance-concerns
Fortune. “Dayton is covering Flock cameras with trash bags after officials found data use violated policy.” 3 June 2026. https://fortune.com/2026/06/03/why-are-ohio-city-workers-covering-flock-cameras-immigration-enforcement-data-sharing-policy-violations/
KATV. “Sherwood stop sparked after Flock camera misread plate by one letter as stolen.” 2026. https://katv.com/news/local/sherwood-stop-sparked-after-flock-camera-misread-plate-by-one-letter-as-stolen
National Today / Morristown. “Flock Safety's AI Cameras Misread Plates, Innocent People Pay.” 9 March 2026. https://nationaltoday.com/us/tn/morristown/news/2026/03/09/flock-safetys-ai-cameras-misread-plates-innocent-people-pay/
Jesse Marx and Dorian Hargrove. “A Flock license plate reader linked a San Diego man to a violent crime. He was five miles away.” Times of San Diego, 7 June 2026. https://timesofsandiego.com/crime/2026/06/07/a-flock-license-plate-reader-linked-a-san-diego-man-to-a-violent-crime-he-was-five-miles-away/
Truthout. “In Atlanta, AI Surveillance Expands as Cop City Reshapes Black Communities.” https://truthout.org/articles/in-atlanta-ai-surveillance-expands-as-cop-city-reshapes-black-communities/
Just Security. “Unpacking the Supreme Court's Chatrie Decision.” 2026. https://www.justsecurity.org/145214/chatrie-fourth-amendment-supreme-court/
Reason. “Judge says warrantless Flock search violated driver's Fourth Amendment rights.” 7 October 2026. https://reason.com/2026/10/07/judge-says-warrantless-flock-search-violated-drivers-fourth-amendment-rights/
Anthony Ha. “Federal judge calls Flock 'indiscriminate mass surveillance'.” TechCrunch, 3 October 2026. https://techcrunch.com/2026/10/03/federal-judge-calls-flock-indiscriminate-mass-surveillance/
Stateline. “Flock Safety tightens safeguards as states, cities question surveillance network.” 14 August 2026. https://stateline.org/2026/08/14/flock-safety-tightens-safeguards-as-states-cities-question-surveillance-network/
Andy Corbley. “US Cities Are Terminating Flock Camera Contracts at Record Pace After Nationwide Pushback.” Good News Network, 31 August 2026. https://www.goodnewsnetwork.org/us-cities-are-terminating-flock-camera-contracts-at-record-pace-in-august-after-nationwide-pushback/
The Texas Tribune. “Texas funding freeze forces cities to address Flock use.” 24 September 2026. https://www.texastribune.org/2026/09/24/texas-flock-cameras-state-funding-local-cities-counties-reaction/
KJZZ. “11 Arizona cities have terminated Flock contracts.” 8 September 2026. https://www.kjzz.org/business/2026-09-08/11-arizona-cities-have-terminated-flock-contracts-one-town-passed-new-rules-to-keep-the-cameras
VPM. “Henrico cancels Flock Safety contract early over 'privacy concerns'.” 6 October 2026. https://www.vpm.org/news/2026-10-06/henrico-police-cancel-flock-safety-alpr-vithoulkas-cooper-newby/
Reuters via Carrier Management. “Flock Safety Will Cut Several Hundred Jobs Amid Surveillance Backlash.” 9 October 2026. https://www.carriermanagement.com/news/2026/10/09/292617.htm
404 Media. “It's not the bots, Flock.” 27 October 2026. https://www.404media.co/its-not-the-bots-flock/
Atlanta News First. “Atlanta mayor to introduce citywide surveillance rules after audit flags 79 APD Flock camera searches.” 15 September 2026. https://www.atlantanewsfirst.com/2026/09/15/atlanta-mayor-introduce-citywide-surveillance-rules-after-audit-flags-79-apd-flock-camera-searches/

Tim Green UK-based Systems Theorist & Independent Technology Writer

Tim explores the intersections of artificial intelligence, decentralised cognition, and posthuman ethics. His work, published at smarterarticles.co.uk, challenges dominant narratives of technological progress while proposing interdisciplinary frameworks for collective intelligence and digital stewardship.

His writing has been featured on Ground News and shared by independent researchers across both academic and technological communities.

ORCID: 0009-0002-0156-9795 Email: tim@smarterarticles.co.uk

Listen to the free weekly SmarterArticles Podcast

Support What We Do

Discuss...

参与讨论