A Field Guide to the Flock

Every product this company sells is named after a bird. Falcon, Sparrow, Raven, Condor. Here is what they actually do, who is really searching them, and what happened when a handful of ordinary people finally read the logs.

← Back to Articles
A Flock Safety automated license plate reader camera mounted on a pole beside its solar panel, photographed against a cloudy sky.
A fixed Flock plate reader. The solar panel is why these go up without an electrician, a trench, or a permit for a power drop. Photo: Tony Webster via Wikimedia Commons, CC BY 2.0. Aurora, Colorado, June 27, 2024.

Verified as of September 9, 2026. Legislative status, contract counts, and litigation posture in this area change week to week. Every claim below carries a date or a named source so you can check what has moved. Where reporting conflicts, both figures appear.

The short version

  • What it is. Flock Safety, founded in Atlanta in 2017, sells solar-powered cameras that photograph every passing vehicle, read the plate, describe the car, and log the time and place. The company reports about 20 billion scans per month across more than 120,000 cameras in 49 states.
  • The load-bearing fact. Flock told Congress that roughly 75 percent of its law enforcement customers opted into the National Lookup Tool, which turns a local camera into a nationally searchable one. Nationwide reach is the default configuration for three out of four agencies, not an abuse of the system.
  • The pattern. Pflugerville, Texas, Mountain View, California, Ventura County, Oxnard, San Francisco, and at least 17 Washington agencies all discovered their data reaching organizations they had not authorized. In most cases a resident or a single council member found it, not the police department and not the vendor.
  • Michigan. More than 180 Michigan agencies use Flock, close to a third of all agencies in the state, up from about 60 in 2022. There is no state law. East Lansing has one camera and its data was searched by dozens of agencies. Ferndale found its own police accounted for less than 1 percent of searches on its own cameras.
  • The AI part nobody explains. Flock accurately says it does not use facial recognition. Its FreeForm feature still lets an officer type a plain-English description of a person and get matching frames. Every rule written around the phrase "facial recognition" misses this entirely.
  • It is not only Flock. On September 2, 2026 Senator Josh Hawley expanded his Senate investigation to Motorola Solutions, Verkada, and Axon Enterprise, asking each the same questions about search scope, retention, misuse, and safeguards. Switching vendors does not resolve any of it.
  • The other side. Flock reports its technology supported more than a million criminal investigations in 2025 and helped locate more than 10,000 missing people. Those are the company's own figures, and they are not nothing.
  • What you can do in an afternoon. Search your plate at HaveIBeenFlocked.com, check DeFlock for cameras near you, and file a records request for your community's Flock network audit. In every case below, the document that changed the outcome already existed. Nobody had asked for it.

The Saginaw Township board of trustees voted four to one on May 29, 2026 to buy 34 automated license plate reader cameras and put them at 15 intersections. The first year runs about $147,000 because of installation. After that it settles at roughly $108,000 a year. Police Chief Scott Malace told the board the cameras would help investigators find stolen vehicles, missing persons, and homicide suspects.

Every word of that is accurate. The cameras do not record video. They take one still photograph of each passing vehicle, read the plate, and log the time and the place. Chief Malace described the product the way it was described to him, and the product does what he said it does.

The vote did a second thing, and that second thing almost never comes up at a township meeting. It enrolled Saginaw Township in a network.

I want to be careful here, because the loose version of this claim is easy to dismiss and the precise version is worse. Flock Safety offers its law enforcement customers several tiers of data sharing, ranging from nothing at all up to a nationwide search network. Moving to the top tier is a checkbox in an administrator panel. Flock told Senator Ron Wyden's office that approximately 75 percent of its law enforcement customers have checked it.

Three out of four. Nationwide reach is not an abuse of the system or an exception to it. For most agencies it is the configuration.

That would be manageable if communities reliably knew which tier they were on. A growing pile of audit records says they often do not.

I write about artificial intelligence for businesses in Midland, Bay City, and Saginaw. I am not a lawyer and I am not going to tell anybody how to vote on a camera contract. I got pulled into this by a narrower question that is squarely my job. What happens when an organization buys a piece of AI software based on a demonstration, and the software turns out to be a platform whose capabilities keep expanding on somebody else's schedule, wired into a network whose scope the buyer never fully understood? That question looks identical in a plant, a marketing department, and a township hall. The Flock story is the most thoroughly documented version of it in the country, and Michigan is far enough into it that the receipts are public.

This is long. The record is long.

1. The machine on the pole

A Flock license plate reader is a solar-powered camera, usually aimed at one direction of traffic, often mounted on a pole that needs no utility connection. That last detail explains a great deal about how fast these multiply. There is no trenching, no electrician, no permit for a power drop. A crew bolts it up and it works.

When a vehicle passes, the camera captures an image and software on the device runs optical character recognition on the plate. This is the same family of technology that reads a check at an ATM or a shipping label at a distribution center, tuned for plates at highway speed and paired with infrared so it works after dark.

The plate is the smallest part of what gets stored. A second model produces what Flock calls a Vehicle Fingerprint, cataloging the car itself: color, make, body style, roof racks, ladder racks, visible damage, bumper stickers. The purpose is to let an investigator search with no plate number at all. A witness who saw a silver pickup with a ladder rack and a dented tailgate has handed police a usable query.

Every record carries a timestamp and a location. Plate, plus vehicle description, plus time, plus place, uploaded to a shared server. One record is trivial. A few billion of them is a different object entirely, and that distinction turns out to be the center of the constitutional fight.

MeasureFigureSource
Vehicle scansAbout 20 billion per monthFlock, cited in Senator Hawley's August 26, 2026 letter
Cameras deployedMore than 120,000 across 49 statesHawley letter; company materials
Law enforcement agenciesMore than 5,000Flock
CommunitiesMore than 6,000Flock
Valuation$7.5 billion in March 2025, reported at $8.4 billion by April 2026Built In; Tech Startups
LobbyingAbout $2 million on federal and state lobbying since 2025, including $840,000 federal in 2025 and $430,000 federal in the first half of 2026. Active lobbyists in 34 states.OpenSecrets analysis, August 2026

Camera counts vary by source and counting method. Earlier tallies put the fleet between 75,000 and 100,000, with reporting that the network crossed 100,000 in June 2026. The 120,000 figure is the one a United States senator put in a formal letter to the company, so it is the one worth citing.

2. Nine years of quiet additions

Flock started in 2017 selling a solar camera to homeowner associations worried about package theft. Laying out what got added to that product, and when, is the clearest way to understand what the word "platform" means in practice.

ProductArrivedWhat it does
Falcon, Sparrow2017 onwardFixed plate readers. Optical character recognition plus Vehicle Fingerprint description. Solar, infrared, no utility hookup.
Raven2021Gunshot detection. Records audio in five-second increments and classifies each clip. Acoustic models reportedly expanded past gunfire to breaking glass.
CondorQ2 2023Pan, tilt, and zoom video camera with people-detection alerts and a Guardian Mode that zooms in on a person and follows them through frame.
FreeFormFebruary 2025Natural-language search over Condor video. An officer types a description and the system returns matching frames.
BravoNot publicly datedEdge compute box that runs classification and event correlation on the device rather than in the cloud.
FlockOSQ2 2023Real-time crime center software unifying plate readers, video, drones, gunshot alerts, dispatch, and records management. Integrates with Axon dash cameras.
Aerodome, AlphaOctober 2024 acquisition; Alpha launched October 2025Drone as first responder. Docked units launch on a 911 call, a plate hit, or a gunshot alert.
NovaAnnounced May 2025Investigative platform joining plate data to commercial people-search records to move from a plate to a person.

The drones

In October 2024 Flock acquired a 17-month-old drone company called Aerodome for more than $300 million. Aerodome had raised $28 million and was founded by former police officer Rahul Sidhu. Both companies are backed by Andreessen Horowitz. Flock is building a 100,000 square foot drone manufacturing facility in Georgia.

The resulting product, Alpha, launched in October 2025. It flies up to 60 miles per hour, carries cameras that Flock says can read a license plate from 2,000 feet, and returns to a dock that swaps batteries and puts it back in the air in under 90 seconds. In January 2025 the Elk Grove, California Police Department became the first agency to receive an FAA waiver letting Flock Aerodome drones fly beyond visual line of sight, without a human observer, up to 400 feet across an entire jurisdiction, with an average response time of about 85 seconds. Plate reads captured in the air feed the same databases and hot list alerts the ground cameras populate.

Nova, and the sentence that gave it away

Internal documents, Slack messages, and meeting audio obtained by 404 Media described Nova as an investigative platform designed to let police, in a Flock employee's own words from an internal meeting, "access data and jump from LPR to person and understand what that context is, link to other people that are related to that person."

The ACLU made the sharpest point. Flock had long maintained that its cameras do not collect personally identifiable information. Nova made that position untenable, because a plate maps to a registered owner through DMV records and the company was now advertising exactly that capability. EFF's Matthew Guariglia put the risk plainly: an officer having someone's entire online persona one click away after a camera scans their plate is an open invitation for retribution based on First Amendment protected expression and affiliations.

404 Media also reported that some Nova data came from breaches rather than public records, including a hacked parking meter app. On May 30, 2025 Flock published a rebuttal titled "Correcting the Record: Flock Nova Will Not Supply Dark Web Data," saying it had explored sourcing that data and decided against it.

The point to sit with. Every capability above was added to contracts communities had already signed. A township that approved a plate reader in 2022 did not vote on drones, on people-detection video, on natural-language person search, or on a fusion platform joining plates to data broker files. Each arrived as a product announcement, an upgrade tier, or a renewal line item. Nobody lied to anybody. The product kept shipping features on the vendor's roadmap rather than the buyer's.

3. The thing that is not facial recognition

Flock is clear and consistent about one boundary. On its trust pages the company states that its cameras do not use facial recognition, that searches are based on vehicle characteristics rather than identity, that the system reports what a vehicle looks like and not who someone is, that it does no predictive policing and no demographic profiling, that searches must relate to a specific investigation, and that local agencies decide where cameras go and who has access.

Take all of that at face value. As far as the public record shows, it is true. It is also, for reasons that take about a minute to explain, close to irrelevant.

Close view of a Flock Safety camera housing mounted on a pole, showing the lens and a ring of infrared LEDs beneath the edge of its solar panel.
The lens and the ring of infrared emitters around it, which is how the camera reads a plate after dark. Flock is accurate that this hardware performs no facial recognition. What the platform does with a typed description of a person is a separate question, and it is the one nobody at a township meeting knows to ask. Photo: Tony Webster via Wikimedia Commons, CC BY 4.0. Jackson County, Colorado, May 25, 2026.

Facial recognition turns a face into a mathematical template and matches that template against a database of known identities. It answers the question "who is this."

There is a second technique that answers a different question, and almost nobody outside the security press has explained it to a general audience. A vision model can be trained to describe a scene in plain language rather than to identify the people in it. It looks at a frame and tags what it observes: a person, an orange vest, a hard hat, a skateboard, tattoos on a forearm, a grey shirt. An officer then types a sentence, and the system retrieves matching frames.

No template. No identity. No biometric anything. The model never knows who it is looking at, and it does not need to, because the officer supplies the name at the end.

What the records actually show

On July 16, 2026, 404 Media published records showing police had used FreeForm to search for people rather than vehicles. Here is where precision matters more than volume, because the honest version of this is smaller than the headlines and still lands.

One county network audit released in full covered 470 FreeForm searches between April 2025 and May 2026. Of those, 456 searched for vehicles. Fourteen searched for people. The text of each query was readable because the releasing agency redacted the searching agency, the officer, the plate, and the case number, but left the text prompt field intact.

New Hanover County, North Carolina, whose network audit produced those records, canceled its Flock contract on September 8, 2026.

The fourteen included "heavy-set male with a black and white hat," "person on skateboard," "person wearing orange vest and construction hat," "male with tattoos," and "white woman wearing grey shirt, blonde hair, black shorts with blue and white shoes." Some searches in the broader body of 404 Media's reporting referenced a target's race or apparent political affiliation.

And here is a detail that belongs in the piece precisely because it cuts the other way. Three separate times, months apart, an officer typed a race descriptor into the search box, and the same search was immediately re-run with the race removed. That looks like Flock's guardrail working as designed. The company says FreeForm alerts administrators when a search includes race, ethnicity, religion, or nationality, and the log appears to show that alert changing officer behavior in real time.

WIRED later collected and analyzed the software and documented a further capability: an officer can draw a boundary around an area of a city and set the cameras inside it to run a continuous, automated search for anyone matching a written description. The interface's own suggested example is "person wearing scrubs." WIRED also surfaced one design detail worth more than a paragraph of argument. When an officer searches for a vehicle, the interface offers dropdown menus for colors, body types, makes, models, roof racks, bumper stickers. When an officer searches for a person, there are no filters at all. It accepts only free text.

Why the denial does not settle anything

If you are trying to find one particular person in a city of 100,000, you do not need to know their face. You need to narrow 100,000 down to a handful. "Heavy-set male, black and white hat, near this intersection, in this hour" does that. An officer looks at the returned images and makes the identification himself. The machine performed the search. The human performed the recognition. The outcome matches facial recognition, arrived at by a route that no existing rule anticipated.

That gap is not hypothetical. A long list of state statutes and municipal ordinances written to restrain police surveillance are built around the phrase "facial recognition." Description search walks past every one of them without technically violating a word, because it never builds the biometric template those rules were written to govern.

When Greenfield, Massachusetts debated its Flock contract, a police lieutenant told residents the cameras show the front or back of a car and nothing more, that there is no angle where you could see someone's face, and that there is no facial recognition and never has been. He was almost certainly telling the truth about his department's plate readers. That is a completely different question from what the platform is capable of, and it is the question nobody at that meeting knew to ask.

4. What you join when you buy one camera

The sharing architecture is the part everyone gets slightly wrong, in both directions. Critics overstate it, and the overstatement gets used to dismiss the whole concern. Departments understate it, often sincerely. The accurate version is more alarming than either.

The tiers, in Flock's own words

One-to-one sharing. A department shares camera access with another specific department. Departments can also share with all agencies inside a set radius, for example everyone within 10 miles, or with all agencies in their state. An agency can share nothing at all, or withdraw at any time.

Statewide sharing. Participating agencies within a state search across that state's network using a full plate. This is a separate setting from one-to-one.

National Lookup. A department opts in, contributes its own data, and in exchange its officers can search every plate captured by every other opted-in camera in the country. Flock describes it as built for far-reaching crimes like human trafficking, narcotics trafficking, and missing persons, and says it permits searching only a complete, known plate rather than partial or descriptive queries.

Flock says a search reason is required every time, chosen from a dropdown plus a free-text case number field, and that every search and its stated reason is preserved permanently in the audit trail of every agency whose camera was included. That last provision is the reason this article is possible at all. The audit trail is how Pflugerville found out.

One checkbox, and one clause

The ACLU of Massachusetts obtained Flock training materials through a statewide records project. The screenshot in those materials shows that enrolling in the national network is a single option reading "Enable National Lookup." The same materials show departments can review individual sharing requests from other agencies, or set the system to accept them automatically.

Wyden's office noted that the 75 percent enrollment figure is likely driven by the structure of the trade. Flock restricts nationwide search access to agencies that agree to share their own data. You get the country, and the country gets you.

The same records project found something further. Flock's default agreement gives the company the right to share data with federal and local agencies for investigative purposes even when a local department restricts data to its own officers. The specific language to look for in a local contract grants Flock "a nonexclusive, worldwide, perpetual, royalty-free right and license" to "disclose the Agency Data (both inclusive of any Footage) to enable law enforcement monitoring against law enforcement hotlists as well as provide Footage search access to law enforcement for investigative purposes only."

Any Michigan community with a Flock contract can pull its own document and search for that phrase. That is a concrete thing a resident or a council member can do in an afternoon.

The terms changed, and the changes point one direction

The ACLU examined Flock's standard terms and conditions dated February 16, 2026 and identified at least four changes made over the preceding seven months.

The previous terms stated flatly that "Flock does not own and shall not sell Customer Data." That sentence is gone. The new license grants formal ownership of the data to the customer while giving Flock "the exclusive right to determine and control the method, timing, format, and medium" of the customer's access to data that supposedly belongs to it. The terms add a perpetual right for Flock to use customer data to support and improve its services, which survives after a city terminates and loses its own access. Disputes go to private arbitration under Georgia law. And termination got harder, including in the specific case where a city council votes not to approve a contract or fails to appropriate the money for it.

Read that last item twice. A city can vote no and still have a harder time getting out.

5. What happens when somebody reads the logs

Pflugerville, Texas

Sam Aly lives in Pflugerville, a city of roughly 68,000 outside Austin. He filed a Texas Public Information Act request for the city's Flock audit records and reviewed about six months of them.

The Pflugerville Police Department made 5,078 searches in that period. Outside agencies made approximately 1,597,918 searches involving a network that included Pflugerville's cameras. That is roughly 315 outside searches for every one run by Pflugerville's own police. The records identified 459 outside organizations, including entities hundreds of miles away inside Texas and organizations outside Texas entirely.

The city had authorized 83 partner agencies, for active criminal investigations only. Flock's own public-facing site listed 78 organizations with access, all of them in Texas. City staff had told residents Pflugerville did not participate in statewide or nationwide sharing.

City Council Member Jonathan Coffman obtained the raw data files, checked Aly's analysis against them, verified it, and posted his confirmation. "We purposely excluded ourselves from any statewide or nationwide databases," Coffman said, "and it turns out that that was not the case."

City Manager James Hartshorn had staff physically cover the lenses of all the city's cameras. On August 25, 2026 the council voted unanimously not to renew. The Pflugerville school district separately pulled Flock cameras from its campuses.

Then two explanations arrived, and they do not match.

Flock spokesman Paris Lewbel said Pflugerville had statewide sharing enabled since 2024, that audit logs confirm it, that agencies control their own sharing settings, and that participating agencies receive monthly emails detailing their sharing relationships, providing additional visibility into how their systems are configured. He said two out-of-state agencies had been mistakenly categorized as Texas agencies, that Flock takes responsibility for that error, and that it was corrected months earlier.

City spokeswoman Ty Bishop said Pflugerville authorized 83 partner agencies for active criminal investigations, and then this: "Flock informed the Police Department about a 'bug' this week that had allowed Pflugerville data to enter national database searches." She added that the monthly dashboard reports Flock provided "did not show this discrepancy and instead showed searches specifically from Pflugerville officers and the City's 83 authorized partner agencies."

Set those side by side. Flock says monthly emails gave the city visibility. The city says the monthly dashboards showed only authorized searches and did not surface the discrepancy. Both cannot be correct.

One caveat Flock is right to raise, and this article should state plainly: a search touching Pflugerville's network does not prove anyone retrieved a Pflugerville plate. The 1.6 million figure measures reach, not retrieval. It is still 1.6 million.

And the part that should worry a Michigan city council is not the number. It is who found it. One resident with a records request found what the police department had not found, what the city council had not found, and what the vendor's own monthly reporting did not surface.

Mountain View, California

An investigation supported by the ACLU found a statewide lookup setting had been active on 29 of Mountain View's 30 Flock cameras since installation, without the police department's knowledge, for 17 months. A separate national lookup setting on one camera had also been toggled on without the city's knowledge.

Over the 12 months ending December 2025, more than 250 agencies that had never signed a data-sharing agreement with Mountain View conducted approximately 600,000 searches of the city's plate data. Among the agencies found accessing it: the Bureau of Alcohol, Tobacco, Firearms and Explosives, the United States Air Force, and the General Services Administration Inspector General.

Police Chief Mike Canfield disclosed this to the city council in February 2026. The council voted to terminate.

Ventura County, Oxnard, and San Francisco

A Ventura County Sheriff's Office audit found out-of-state agencies had accessed its plate reader data more than 364,000 times without approval or knowledge. The National Lookup feature had been disabled in June 2023 specifically to comply with California law restricting the sharing of plate data. It was active again anyway. "No one from our agency activated the national lookup feature," the Sheriff's Office said.

Flock offered three possible explanations: an employee from the sheriff's office, a Flock employee, or an automatic system bug. The company said technical logging limitations prevented it from determining which of the three it was.

That sentence deserves to be read twice. A vendor that permanently logs every customer search, and holds that up as its core accountability mechanism, could not determine who changed a customer's single most consequential setting.

The Ventura Police Department ran its own 90-day audit and identified a vendor-based configuration error on February 23, 2026. It added daily verification checks. The Oxnard Police Department suspended its fixed Flock cameras after a routine audit revealed a vendor-based issue had silently re-enabled the nationwide query feature. A San Francisco Police Department audit showed out-of-state and federal agencies queried the city's Flock database more than 1.6 million times over seven months, a figure that now anchors a class action.

Washington State, and the three doors

The University of Washington Center for Human Rights conducted the most systematic review of any state's Flock records and published it on October 21, 2025 under the title "Leaving the Door Wide Open." Its taxonomy is the most useful thing anybody has produced on this subject.

Front door. At least eight Washington agencies enabled direct one-to-one sharing with U.S. Border Patrol at some point during 2025.

Back door. Audits revealed apparent Border Patrol access to the networks of at least ten Washington police departments that never explicitly authorized it.

Side door. Audits revealed local agencies running searches on behalf of federal immigration enforcement.

In total, Border Patrol searched the plate reader data of at least 17 Washington agencies, including sheriff's offices in Benton and Yakima counties and police departments in Auburn, Black Diamond, Centralia, Chehalis, Eatonville, Ellensburg, Grandview, Lakewood, Mukilteo, Prosser, Renton, Richland, East Wenatchee, Wenatchee, and Yakima. The researchers also found broad discrepancies between what network audits contained and what agencies' own "Networks I'm Sharing With" reports and transparency portals listed.

Milwaukee, reaching through Illinois

404 Media published a data trove derived from Flock audits obtained from the Danville, Illinois Police Department. The Wisconsin Examiner searched it and found the Southeastern Threat Analysis Center, a homeland security arm of the Milwaukee Police Department's fusion center, in the records.

On July 15 and October 21, STAC personnel ran three searches inside Danville's Flock network. The July 15 search alone reached 4,893 Flock networks and an equal number of individual devices. The October searches reached 5,425 networks and pulled a month of data. Milwaukee police said the investigation was classified and ongoing, related to a Homeland Security Investigations case under a federal drug trafficking task force, and not immigration related.

An agency in Wisconsin, searching through a network in Illinois, touching cameras across nearly five thousand jurisdictions, none of which had any relationship with Milwaukee.

6. Michigan, in detail

The scale, and the speed

Michigan has no state law governing automated license plate readers. That single fact organizes everything below, because it means every decision belongs to a township board, a city commission, or a county board of commissioners, made in most cases by people seeing a vendor presentation for the first time.

More than 180 Michigan law enforcement agencies use Flock, close to a third of every agency in the state. In 2022 the number was roughly 60. Bridge Michigan counted at least 125 county or municipal agencies under contract, with cameras in use in 50 of Michigan's 83 counties. The crowdsourced DeFlock database lists roughly 3,557 reported reader locations statewide, clustered around metro Detroit, Battle Creek, Flint, Grand Rapids, Holland, Jackson, Kalamazoo, Lansing, Muskegon, Saginaw, and Traverse City. That count ranks Michigan eighth among states and territories. DeFlock entries are community-maintained metadata contributed by volunteer mappers through OpenStreetMap, not proof of an active contract, and anybody citing them should say so.

Tripling in four years, with no statute, is the actual Michigan story.

East Lansing already has the Pflugerville document

Michigan does not have to wait for somebody to file the request. Somebody already did.

A 1,700-page release obtained through a records request and reported by East Lansing Info shows the East Lansing Police Department's Flock database being searched by dozens of police departments around Michigan across a 30-day stretch in July and August.

East Lansing officers searched the database fewer than 200 times in that window. Their searches came in sporadically. At one point an officer ran more than 30 searches in a 12-minute stretch. Across one six-day span in August, East Lansing officers ran no searches at all.

Meanwhile dozens of other agencies searched East Lansing's data: departments across Metro Detroit, several sheriff's offices, the Gogebic County Sheriff's Office in the westernmost county of the Upper Peninsula, and most frequently the Michigan State Police. The Michigan Department of Natural Resources, whose mandate is conservation, also tapped the city's Flock data.

The reach numbers are the part to sit with. East Lansing's own searches touched between 1 and 855 networks, with most including more than 800. Michigan State Police searches frequently reached more than 6,000 networks.

East Lansing has one Flock camera.

One camera on one pole in one city, feeding a database searched by agencies across the state, reached by queries spanning thousands of networks, operated by a police department that went six days without running a single search of its own. The Michigan State Police position on its own use is that all use of the system is restricted to legitimate law enforcement purposes, that searches and access are logged and auditable, and that the technology is not used for generalized surveillance.

Ferndale read its own logs and left

Ferndale ended its Flock contract on November 13, 2025 after an internal audit conducted by a city council member.

The audit found the Ferndale Police Department accounted for less than 1 percent of all searches run through Ferndale's own cameras. The audit log, covering September 2024 through September 2025, showed the top searchers were Dallas and Houston police. Houston accessed Ferndale's data more than 10,000 times in a single month. Ferndale itself topped 1,000 searches in a month exactly once.

Customs and Border Protection, Homeland Security, and two regional ATF offices also appeared in the log. The records surfaced hundreds of searches using terms related to immigration, graffiti, anti-Trump protests, and littering.

Work through the arithmetic. A Michigan city bought cameras, mounted them on its own streets, paid for them with its own residents' money, and produced a data stream its own police used for less than one search in a hundred. The overwhelming majority of value generated by Ferndale's investment went to agencies accountable to nobody in Ferndale, for purposes that included graffiti enforcement, littering enforcement, and the monitoring of political protests.

That is not an allegation from an advocacy group. It is a council member reading his own city's logs.

Lansing, and a program with no vote on record

The Lansing Police Department has operated Flock cameras since September 2025, 26 of them as of August 2026. No public record shows the Lansing City Council voted to create the program. When a council committee reviewed it for the first time in April 2026, the meeting packet contained no contract, no written policy, and no audit.

On August 24, 2026 the council voted 7 to 1 on a resolution introduced by 1st Ward Council Member Ryan Kost asking the administration and the police department to end all Flock contracts. The resolution is non-binding. The mayor has to approve any actual termination.

Kalamazoo counts the searches

Kalamazoo produced the most concrete Michigan usage figure available. MLive reported that the 39 readers operated by the Kalamazoo Department of Public Safety were searched more than 750 times per day on average, and that seven municipal, county, or state agencies can reach them.

The city's response is worth copying. On August 17, 2026, after several dozen residents turned a commission meeting into a protest over a camera system that was not even on the agenda, the Kalamazoo City Commission voted unanimously to require commission approval for every Flock contract, regardless of whether the cost falls below the threshold that normally lets staff sign without a vote.

That is the quiet mechanism at the center of this whole story. These contracts are often small enough to be approved administratively, which is how a surveillance network gets built without a single public vote. Kalamazoo did not ban anything. It just moved the decision back into the room where the public can watch it.

The Tri-Cities split

Saginaw and Flint use the cameras. Saginaw Township approved its 34 on a four to one vote in May 2026.

The Bay City Commission went the other way. On November 17, 2025 it voted 6 to 2 to reject a two-year contract for 13 cameras at a total cost of $84,150, after months of public debate and objections from the ACLU. Resident Adam Reddick pointed commissioners at contract language allowing Flock to access, use, preserve, and disclose customer data to government officials and law enforcement. There are now no publicly operated plate reader cameras in Bay City.

Two communities in the same media market, the same vendor, the same pitch, opposite conclusions, and no state standard telling either one which answer is correct.

Saginaw Township then did something worth naming, because it is the most constructive response any Great Lakes Bay Region community has made. On August 24, 2026 the township police department launched a public Flock transparency portal showing when searches were conducted, what offense type each was tied to, and how many cameras were searched. Lieutenant Andy Myers said the portal is meant to demonstrate the department is using the system properly. Three days earlier, on August 21, the department credited Flock data with identifying two people who had been hitting multiple stores in the area.

A portal is not an audit of who outside Saginaw Township is searching Saginaw Township's cameras. That is a different document, and it is the one Pflugerville and Ferndale had to request. But publishing anything voluntarily puts Saginaw Township ahead of most of the state, and the residents who wanted accountability now have a place to start.

Waterford, Oakland County, and how these systems grow

Waterford Township began using plate readers in 2022 and now runs 16 of them, paid for entirely out of federal drug forfeiture funds at a little over $160,000 since January 2022. In August 2025 trustees approved a separate three-year, $399,000 contract for three Flock Aerodome drones. In February 2026 they added a three-year, $60,000 video integration agreement covering Flock OS-Plus, which joins Flock footage to video from cruiser dash cameras, body cameras, and the Waterford School District.

A Flock Safety installation on an existing green streetlight pole: a solar panel on a bracket, a multi-lens dome unit mounted beneath it, and a separate plate reader strapped lower down the pole.
One pole, one solar panel, two devices, strapped to a streetlight that was already there. This is what accretion looks like in hardware. The mount, the power, and the contract are shared, so the next module is an add-on rather than a new vote. Photo: Bruxton via Wikimedia Commons, CC0.

Nobody in Waterford voted for a system linking street cameras to school district video and an aerial drone fleet. They voted for license plate readers in 2022. Every expansion since arrived as its own agenda item.

Oakland County is the sharper example. In April 2026, after a public comment period that ran roughly six hours and was dominated by opponents who booed and taunted commissioners, the Board of Commissioners approved a nine-month free drone-as-first-responder pilot for the Sheriff's Office called Project Prove It, providing seven Flock drones and supporting equipment.

Under the original contract language, a two-year, $2.5 million deal would have kicked in automatically unless the county opted out. An amendment passed 16 to 1 requiring the full commission to approve any post-pilot contract, mandating that all data collected belongs to the county, and restricting drone use to calls for service, deputy requests, and search-and-rescue.

A free pilot that converts to a $2.5 million contract by default is not a hidden trap. It is a standard commercial term, printed in the agreement, and it is exactly the kind of clause that gets skimmed when the discussion is about drones and the crowd is loud. Oakland County caught it. That amendment is the single most transferable thing in this entire article for anybody who signs vendor contracts.

The document you may not be able to get

Josh Tewolde is a student at Kettering University in Flint. He filed a Michigan Freedom of Information Act request with Genesee County asking for a map of camera locations, audit logs identifying who accessed the system and when, and records of scans of his own vehicle.

The county released its Flock contract. It denied the camera locations and the audit logs, explaining in part that although it had a contract with Flock, it did not own the system or the data within it.

Tewolde appealed, citing a line from the county's own contract: "As between Flock and Agency, all right, title and interest in the Agency Data, belong to and are retained solely by Agency." The county denied the appeal, saying the audit and search logs did not exist within its files.

The Mackinac Center Legal Foundation filed suit on his behalf on August 18, 2026 in Genesee County Circuit Court. Senior attorney Steve Delie framed it in one line worth quoting: "Genesee County can't outsource transparency to a private vendor."

Kent, Washington shows where the county's argument leads if it succeeds. When University of Washington researchers requested Kent's Flock network audit, the city attorney determined it was not a public record under the Washington Public Records Act. The reasoning: the network audit is automatically generated by Flock using searches from other agencies, Kent does not contribute to its preparation, the agreement distinguishes data provided by the city from data owned and aggregated solely by Flock, and the police department does not retain the audit because it lives entirely on Flock's servers.

Follow that logic to its end. The single document that would reveal who has been searching your community's cameras may not be a public record in your community, because your city did not create it and does not hold it. Pflugerville got its answer. Kent did not. Genesee County is in court over the same question.

Who has left

Michigan communities that have canceled, declined to renew, or voted down Flock agreements include Bay City, Ferndale, Westland, Clawson, Milford, Rochester, Sumpter Township, Menominee, Oceana County, and Dickinson County. In Westland, Police Chief Kyle Dawley simply declined to bring the expiring contract before the City Council for renewal, and the cameras were gone by the end of July.

In Lapeer County, retired attorney Linda Berker told commissioners weighing a sheriff's department purchase: "I see this as one big, slippery slope. I know we have a right to safety, but what about privacy, and what about the right to go wherever you want, whenever you want, without the government tracking you?"

Gabrielle Dresner, a policy strategist with the ACLU of Michigan, on the 30-day retention standard that was the industry norm: "That's an extensive amount of travel data that's being held on people just traversing the roadways, doing nothing wrong. You can see a pattern of people's movements based on where their license plate crosses through."

Three bills, and a governor's race

On January 29, 2026, during Data Privacy Week, Representatives Doug Wozniak, a Republican from Shelby Township, and Jimmie Wilson Jr., a Democrat from Ypsilanti, introduced House Bills 5492 and 5493. The bills are tie-barred, so neither takes effect unless both are enacted. They were referred to the House Committee on Judiciary, and the House Fiscal Agency published its analysis on April 22, 2026.

Wozniak's HB 5492 governs private operators of reader systems. Wilson's HB 5493 governs state and local government entities, and it is the tighter of the two. It would limit government use of readers to comparing captured plate data against alert data for uninsured, unregistered, or stolen vehicles and for electronic toll collection. Captured data could not be used or shared for any other purpose, could not be retained beyond 14 days absent a warrant, court order, or preservation request, and agencies would have to adopt and publish usage policies and file quarterly reports including the number of plates scanned and the number of matches that produced an arrest.

In August 2026, Senator Jim Runestad, a Republican representing Michigan's 23rd Senate District, introduced Senate Bill 1131. It would cut retention from 30 days to 14 and require a warrant for access to reader databases for anything beyond active criminal investigations, missing persons cases, and stolen vehicle investigations. It would also require a registered vehicle owner's consent before information about their car could be released through a Freedom of Information Act request, which cuts directly at the question Genesee County is litigating.

For context on where Michigan sits: 21 states have an automated license plate reader statute on the books, among them Illinois, Virginia, Washington, California, Maine, Montana, New Hampshire, and Vermont. In 28 states there is no retention limit in law at all, which means a photograph of a car can sit in a database indefinitely. Michigan is in the second group.

Michiganders deserve to know that new technology is being used responsibly, not in ways that invade privacy or erode public confidence. Rep. Doug Wozniak, R-Shelby Township
We can support effective policing while still demanding safeguards that protect civil liberties. This legislation creates clear limits on how ALPR data is collected, stored, and shared, ensuring these tools are used to improve public safety, not to enable routine mass surveillance. Rep. Jimmie Wilson Jr., D-Ypsilanti

A Democrat from Washtenaw County and a Republican from Macomb County wrote the House package together, and a Republican from Oakland County wrote the Senate bill. Anybody trying to sort this issue into a familiar political shape is going to have a hard time.

It reached the governor's race, too. Both major candidates have taken positions. Jocelyn Benson has called for stricter oversight, arguing Michiganders deserve basic transparency about where their data goes and who accesses it, and has said she would work with the Legislature to codify rules on camera use, data access, sharing limits, and retention. John James has said "balance is the key," told the Washington Examiner that "our Bill of Rights, our Fourth Amendment, our right to privacy is nonnegotiable," and said he would protect the privacy of Michiganders while also enforcing the law.

When a surveillance vendor becomes a campaign issue in a governor's race, the local-decision-only era is ending. What replaces it is still open.

7. Misuse, error, and the plumbing

Officers using the network on women they know

The Washington Post published an investigation on August 2, 2026 that reviewed police and court records and identified at least 50 law enforcement officers across the United States charged with or accused of using plate reader technology for unauthorized purposes. Flock's system was involved in 46 of the 50. The rest involved competing products from Axon and Motorola. In 26 of the cases, investigators or prosecutors said officers used the technology to surveil wives, girlfriends, ex-partners, their exes' new partners, or women they wanted to meet.

Josue Ayala, an eight-year veteran of the Milwaukee Police Department, was charged after prosecutors said he ran 179 searches for personal reasons, more than 50 of a person he was dating and more than 100 of that person's ex-partner. Milwaukee policy required a case number and an investigative purpose for every search. Prosecutors say Ayala entered a single word each time. He received probation and a $500 fine. A Milwaukee internal affairs detective was later charged with misconduct while investigating Flock system misuse.

Jarmarus Brown, a police officer in Orange City, Florida, ran tags for three vehicles repeatedly over seven months, placed a GPS tracker in his former girlfriend's wallet, and showed her footage from Flock cameras. He faces charges of accessing a computer or electronic device without authority and misdemeanor stalking. A police affidavit obtained by 404 Media says a colleague sitting in an adjacent cruiser looked over and saw a plate reader image of Brown's ex-girlfriend on his screen.

Michael Steffman, the police chief in Braselton, Georgia, resigned and was arrested by the Georgia Bureau of Investigation on November 19, 2025 on harassment and stalking charges, after women seeking protective orders accused him of using Flock cameras to follow them and send them maps of their own movements. Marci Bakely, a 50-year-old single mother, told the Washington Post that when she went to buy groceries or on a date or to the doctor, he would text her about it, sometimes within minutes. Audit logs compiled by HaveIBeenFlocked show her plate and her teenage daughter's plate searched roughly 600 times. Steffman died in April 2026 before the case resolved.

In Stow, Ohio, patrolman Jason Rogers was placed on unpaid leave and criminally charged after allegedly using Flock and other databases to stalk a former partner who said she had changed her identity to escape his abuse. In Lufkin, Texas, officer Zachary Klein was arrested and indicted on 100 felony counts. In California, former deputy Alexander Vanny was convicted on 10 charges including stalking and unauthorized access.

Read enough of these and a pattern emerges that has nothing to do with license plates. Almost none were caught by internal review. Ayala's misuse surfaced because a victim looked up her own plate on a public records aggregator and then called a different police department. Rogers was discovered the same way. The Institute for Justice noted that only a handful of the dozens of cases it analyzed were initially found through internal investigation.

An audit log nobody reads is not an audit. It is a record of what happened, available afterward, to whoever thinks to look.

The counterexample, which matters. On August 24, 2026 the Georgia Bureau of Investigation announced the arrests of four former Savannah Police Department employees for accessing Flock for non-law-enforcement purposes. Three former officers were charged with two counts each of misuse of a license plate reader system and violation of oath of office. A former community service specialist was charged with two counts of misuse. They were among six employees fired on August 7. The GBI said the case originated from an internal audit by Savannah police.

St. Charles County, Missouri is the sharper version, and it is the case that put Hawley onto this. Flock briefed Police Chief Kurt Frisz on its new Audit Assistance feature on August 12, 2026. Two days later, on August 14, the audit flagged an abnormal pattern of searches by a civilian employee with database access. The employee admitted using the system to keep tabs on an ex, was suspended, placed on unpaid leave, received a termination notice, and resigned. On August 20 the county canceled its Flock contract and put plastic bags over its 28 cameras.

Read that sequence carefully, because it cuts both ways and the article would be dishonest to use only half of it. The tool worked. Flock's audit feature caught a stalker in 48 hours, which is 48 hours rather than the months or years it took a victim with a records aggregator in most of the cases above. And the county canceled anyway. Capability and trust are different things, and a vendor can win the first while losing the second.

Auditing works. It works when somebody reads it.

The system used for things it was not sold for

Immigration. In August 2025 the Illinois Secretary of State's office audited a sample of just 12 local agencies. It found U.S. Customs and Border Protection had gained direct access to Illinois camera networks, in violation of a 2023 state law barring plate data from being used to investigate immigration status or out-of-state abortion access without a court warrant. Secretary of State Alexi Giannoulias announced the findings on August 25, 2025, ordered access shut off, and called on departments statewide to re-examine their agreements. His office said Flock did not have proper safeguards in place, and that the problem was compounded by a CBP pilot program the company's own leadership was unaware of. A company was running a federal pilot inside its own product without its executives knowing.

Flock says it ended its limited pilots with CBP and Homeland Security Investigations on August 25, 2025, that it has no contract with ICE, does not sell to or share data with ICE, and that federal organizations were removed from statewide and national lookup in August 2025 and limited to one-to-one relationships. The Atlanta Press Collective found the final direct CBP search hitting Atlanta's network came the day before that cutoff, and that other local agencies continued running searches on behalf of CBP afterward.

Reproductive health. On May 9, 2025, a deputy with the Johnson County, Texas Sheriff's Office queried the network. The reason entered in the search log was "had an abortion, search for female." Records obtained by EFF show two searches, the second reaching 6,809 networks made up of 83,345 cameras across a full month, including cameras in Washington and Illinois where abortion is legal and where laws exist specifically to prevent that kind of search.

Sheriff Adam King and Flock both said publicly the search was a welfare check for a missing person, and Flock called the reporting false and clickbait. Documents and court records EFF obtained afterward show deputies had opened a death investigation of a non-viable fetus, put up crime scene tape, collected evidence, consulted the district attorney's office about whether the woman could be charged, and ran the plate search two and a half hours after the initial call. The investigating officer's own affidavit contradicts key claims used to support the welfare framing.

Protests. EFF obtained datasets representing more than 12 million searches logged by more than 3,900 agencies between December 2024 and October 2025. More than 50 federal, state, and local agencies ran hundreds of searches connected to protest activity, in some cases targeting named activist groups, spanning the 50501 protests in February, the Hands Off protests in April, and the No Kings protests in June and October. Separately, 404 Media reported on a Flock webinar in which the company's director of market management walked police through using FlockOS real-time crime centers to monitor "everything from small parades, to unexpected protests, to unprecedented disasters."

None of that was a hack. None of it was a bug. Every one of those searches was a licensed user typing into a supported feature.

When the machine is wrong, and when the database is

Accuracy here is not one number. It is at least three, and they measure different things. Getting this right matters, because the sloppy version of this argument is the easiest thing in the article to knock down.

What is measuredFigureSource and caveat
Character accuracyMore than 96 percent of plate characters read correctlyFlock marketing. Characters, not whole plates.
Plate accuracy93 of every 100 platesFlock, stated in litigation. At 20 billion scans a month, that implies roughly 1.4 billion misreads monthly.
Alert accuracy, misreads71 percent of 1,427 alerts contained misread platesRoseville, California police department's own internal analysis across 2023 and 2024.
Alert accuracy, bad data161 of about 498 stolen-vehicle alerts wrong, roughly one in threeLAPD Office of Inspector General, July 10, 2026. Most errors traced to stale hot list entries, not camera misreads.

That last row is the one most coverage gets wrong, and it is the most instructive. The Los Angeles audit examined alerts from August and September 2025. It found 161 vehicles falsely flagged as stolen against 337 accurate flags. But the cameras had mostly read the plates correctly. The databases were wrong. Recovered vehicles were never cleared from hot lists, theft reports were outdated, and data entry errors were never corrected. Los Angeles let its Flock contract expire in July 2026.

That distinction is the whole ballgame for anybody deploying AI in a business. Roseville's problem was model accuracy. Los Angeles had a data hygiene problem wearing an AI costume. The camera did its job and handed a correct answer to a system holding stale records, and the output was an armed traffic stop of an innocent person. If your team is evaluating an AI vendor, the model's accuracy is only half the question. The other half is the quality of whatever it is querying, and almost nobody audits that.

The Institute for Justice has documented at least 27 cases since 2018 of innocent motorists pulled over, detained at gunpoint, or jailed after plate reader errors, most of them since 2023. In nearly two-thirds of those cases, officers did not realize the error until after they had drawn and pointed their guns. IJ notes it has identified exactly one jurisdiction in the country that publishes regular reports of such errors, Oak Park, Illinois, which means 27 is almost certainly an undercount.

In Aurora, Colorado in 2020, Brittney Gilliam and four children ranging from 6 to 17 years old were held at gunpoint and several were handcuffed face down on hot pavement. The reader had matched her Colorado SUV plate to a stolen Montana motorcycle plate. An SUV is not a motorcycle, and the mismatch was visible to any officer who looked, but no verification step existed. The city settled for $1.9 million in February 2024.

In Española, New Mexico, a camera read a 2 as a 7 and officers handcuffed a 12-year-old. In Toledo, Ohio, a 7 read as a 2 led to a gunpoint stop and a police dog deployment. In Sherwood, Arkansas, a misread digit flagged a couple's SUV as stolen and officers ordered them out at gunpoint while their six-week-old baby stayed in the back seat. In Plymouth, Minnesota, automotive journalist Joel Feder was tracked for two days in a loaner Range Rover with manufacturer plates and then boxed in by four squad cars outside a Kohl's, traced to a federal database entry missing two middle digits that had been printed smaller than the rest.

Strip away the specifics and every one of those has the identical structure. A statistical model produced a confidence-weighted guess. A human treated the guess as an established fact. No verification step existed between the alert and the armed encounter.

The plumbing underneath

On November 3, 2025 Senator Ron Wyden of Oregon and Representative Raja Krishnamoorthi of Illinois wrote to the Federal Trade Commission asking it to investigate Flock for negligent cybersecurity. Their letter states that Flock does not require multi-factor authentication for law enforcement accounts, continues to support text-message-based MFA vulnerable to interception and phishing, and does not natively support the phishing-resistant MFA the federal government treats as the standard. They cited a public site operated by the security firm Hudson Rock reporting that passwords for at least 35 Flock customer accounts had been stolen by hackers.

Their argument turns on the 75 percent figure. Because most agencies are enrolled in national lookup, anyone who compromises a single law enforcement account can search the country. The lawmakers spelled out what the underlying data reveals about a person: trips to doctors and therapists, support group meetings for alcohol or drug addiction, places of worship, and protests.

The account layer is not the only exposed surface. Independent security researcher Jon Gaines, who publishes as GainSec, formalized a white paper in November 2025 documenting 51 findings across Flock's Falcon and Sparrow plate readers, its Raven gunshot detector, and its Bravo edge compute device, with 22 assigned CVE identifiers and 8 more pending at publication. The categories are the ordinary embarrassments of hardware built fast: credentials shipped in the clear, unauthenticated administrative endpoints reachable from the same network, and paths to root-level access on a device with brief physical contact. The specifics are published and are not worth repeating here, but the shape of the finding is worth stating plainly. These are poles on public streets, in parking lots, at the edges of neighborhoods, and physical access to them is not hard to get.

8. Courts, Congress, and two governors

The courts

In October 2024, Norfolk resident Lee Schmidt and Portsmouth resident Crystal Arrington, a home health care worker who drives clients through Norfolk, sued the city with backing from the Institute for Justice. Norfolk had installed roughly 176 Flock cameras beginning in 2023. Schmidt's and Arrington's vehicles were captured 475 and 325 times respectively over four months in 2025.

A Flock Safety camera on a slim pole with a small solar panel, standing in the grass verge of a two-lane suburban road lined with utility poles and trees, with cars in the distance.
A reader on an ordinary arterial road in Round Lake Beach, Illinois. The Fourth Amendment cases turn on exactly this setting: a camera in a public place, photographing cars that anyone standing on the verge could see. The plaintiffs' argument is not that a single photograph is private. It is that enough of them, kept long enough, become something else. Photo: AlphaBeta135 via Wikimedia Commons, CC BY 4.0. July 23, 2026.

The case swung both directions. In February 2025 Chief Judge Mark S. Davis of the Eastern District of Virginia denied the city's motion to dismiss, writing that a reasonable person could believe society's expectations as laid out in Carpenter v. United States were being violated. In June 2025 he denied Flock's own belated attempt to intervene, writing that the company "made a conscious gamble to not show up to the platform on time."

Then on January 27, 2026 Davis granted summary judgment for the city. In a 51-page opinion he found the plaintiffs could not demonstrate the system was capable of tracking the whole of a person's movements, pointing to Norfolk's 21-day retention, the maximum Virginia law allows, and gaps in coverage. He also wrote that the answer might not hold: "Because rapid technological advances, such as the rise of artificial intelligence, make it impossible to predict how police surveillance will evolve, the Fourth Amendment analysis must remain nimble." And more pointedly, that at some point plate reader surveillance could become too intrusive, but that "at least in Norfolk, Virginia, the answer is: not today."

The plaintiffs appealed. The case is fully briefed at the Fourth Circuit, and as of September 9, 2026 no oral argument has been scheduled and the court has not ruled. The ACLU, the ACLU of Virginia, and the Electronic Frontier Foundation filed a joint amicus brief supporting the plaintiffs on April 20, 2026, and the Cato Institute filed separately. The Institute for Justice has filed a parallel challenge on behalf of San Jose, California residents.

The ground may have shifted underneath all of it. On June 29, 2026 the Supreme Court decided Chatrie v. United States, holding 6 to 3, with Justice Kagan writing for the majority, that law enforcement conducted a Fourth Amendment search when it obtained geofence location data, because an individual retains a reasonable expectation of privacy in location history even when a third party holds it. The Court did not decide whether the particular warrant satisfied probable cause and particularity, and remanded to the Fourth Circuit. Institute for Justice attorneys expect the reasoning to shape plate reader litigation directly.

Flock's counter deserves quoting rather than paraphrasing: "More than 40 courts across the country have considered LPRs in the context of the Fourth Amendment and ruled overwhelmingly that the use of these tools in public spaces does not require a warrant. Rather than banning a critical public safety tool, Congress should consider strong, clear safeguards that ensure accountability, oversight, and responsible use are enshrined in law." Independent reporting puts the verifiable count at appellate and federal district courts in at least 14 states plus the Ninth and Eleventh Circuits, which supports the direction of Flock's claim if not the exact number.

There is also a class action. On February 26, 2026 the Oakland firm Gibbs Mura filed suit in San Francisco Superior Court alongside Milberg PLLC, alleging Flock violated California's ALPR Privacy Act by sharing millions of Californians' movements with out-of-state and federal agencies. The complaint cites more than 1.6 million queries of San Francisco's database over seven months and seeks $2,500 per violation under California Civil Code section 1798.90.55. Gibbs Mura previously secured a $27.5 million settlement against Thomson Reuters over its CLEAR platform and the $1.5 billion Equifax data breach settlement.

Congress

On August 26, 2026, Senator Josh Hawley, Republican of Missouri and chairman of the Senate Judiciary Subcommittee on Crime and Counterterrorism, opened a formal investigation into Flock's collection, retention, and dissemination of data. He demanded documents by September 8, 2026, including a list of any known or suspected misuse of Flock systems since 2021 and files on camera accuracy, security, and what data beyond plates the cameras collect. He cited the St. Charles County case in his own state, and wrote that Missouri is not an outlier.

Congress never authorized the network your industry has built. Your company's internal policies are, in practice, the only safeguards for hundreds of millions of Americans. People across the country are now discovering that fact and demanding answers from their elected officials. Sen. Josh Hawley, R-Missouri, letter to Flock CEO Garrett Langley, August 26, 2026

Flock spokesman Paris Lewbel told Axios the company had received the letter and appreciated "the opportunity to provide the Subcommittee with additional information about Flock and our technology," and said Flock intends to cooperate. Whether the company met the September 8 deadline in full has not been made public.

On September 2, 2026, before that deadline arrived, Hawley widened the inquiry past Flock entirely. He sent letters to the chief executives of Motorola Solutions, Verkada, and Axon Enterprise, giving each until September 16 to answer the same core questions: how broadly customers can search vehicle data, how long the company retains plate information, how many confirmed cases of misuse have occurred since 2021, and what safeguards require an officer to give a legitimate justification before searching. His framing was that the problem is structural rather than corporate: "Recent examples of abuse and misuse of personal data reveal the woeful lack of safeguards in these systems."

Of the three, only Axon commented publicly in the week that followed. A spokesperson said the company had received the letter, had reached out to Hawley's office to discuss the requests, and welcomed the opportunity to demonstrate its commitment to safety and responsible innovation. Motorola Solutions and Verkada had not publicly commented as of September 9, 2026, which is unremarkable a week before a deadline and should not be read as refusal. Congressional document productions are ordinarily confidential, so whether Flock met its own September 8 deadline in full is also not public.

That expansion matters more than the original letter. It is the first formal signal from Congress that this is a category of technology to be governed, not one vendor to be scolded. Any community that cancels a Flock contract and signs with a competitor, as Ferndale did when it moved to Axon, is buying into the same unresolved questions, from a company now being asked the same four.

From the other side of the aisle, Krishnamoorthi and Representative Robert Garcia, ranking member of the House Oversight Committee, opened an investigation in August 2025 after reports the system had been used to track women across state lines following abortion care and to conduct unauthorized immigration enforcement.

For all that attention, no bill in Congress would ban or comprehensively regulate this category of surveillance. An OpenSecrets analysis published in August 2026 found Flock has spent about $2 million on federal and state lobbying since 2025, including $840,000 federally in 2025 and $430,000 in the first half of 2026, with active lobbyists in 34 states. Its April through June 2026 quarter, at $220,000, was its highest since it began federal lobbying in 2024.

Two governors, four days apart

In late August 2026, Texas Governor Greg Abbott ordered state agencies to stop spending public funds on Flock cameras. Reporting places the directive on August 26, 27, or 28 depending on the outlet. EFF reported the directive arrived just as The Texas Tribune prepared to publish an investigation showing a state agency had quietly funneled at least $30 million into building the network. The Motor Vehicle Crime Prevention Authority, whose board is mostly Abbott appointees, helped install at least 3,200 Flock cameras since 2023, funded by a 2023 law that added $1 to every Texas auto insurance policy to combat catalytic converter theft.

Abbott did not order removal, saying most funding comes from the federal government and it is up to city officials to unplug. The Texas Department of Public Safety, which had at least 940 cameras installed, said on September 4 that it will keep operating them for what it called legitimate public safety efforts. It did not explain how it would fund recurring contracts without state money. After questions about who could reach its network, DPS said it would begin limiting access to agencies holding a signed agreement with DPS, which is the same fix Pflugerville thought it already had.

Days later, following an August 26 press conference where Governor Ron DeSantis called plate reader use "out of control" and said he does not want Florida to become a surveillance state, the Florida Department of Transportation issued a memorandum on August 31, 2026 directing local law enforcement to remove all license plate readers from state highway rights-of-way within 30 days, a deadline of September 30. FDOT said it will remove any that remain itself and will stop approving future installation requests. The department cited an exponential increase in deployments alongside reports of misuse and data privacy concerns. The order covers all plate readers on state rights-of-way, not only Flock's.

How many communities have left

Counts vary by source and counting method, which is itself worth saying rather than picking a favorite number. Cybernews reported 214 cities and towns have dropped Flock since 2021, with 90 canceling in August 2026 alone, a fourfold increase over July. DeFlock counted at least 56 communities that deactivated, rejected, or canceled during 2026, spanning 23 states, with California and Wisconsin at seven each. Newsweek has mapped 149. The Institute for Justice maintains its own database.

The first week of September did not slow it down. Both Fond du Lac County law enforcement agencies and the Winnebago County Sheriff's Office in Wisconsin suspended all Flock products immediately. Sheboygan's Common Council voted unanimously to terminate, keep the cameras covered until the contract ends, and strike the appropriation from its 2027 budget. New Hanover County, North Carolina canceled. Cave Creek, Arizona, Bellingham, Massachusetts, and several Texas counties ended their use. Whatever the running total is on the day you read this, it is higher than the one printed here.

The direction is unambiguous even where the arithmetic is not.

9. Flock's case, made properly

An article like this is worthless if it presents one side, and Flock has a real one.

The company reports that in 2025 its technology supported more than one million criminal investigations and public safety incidents, assisted in an estimated 20 percent of solved cases in jurisdictions where it was deployed, and helped locate more than 10,000 missing people, an average of more than 27 per day. More than 40 percent of customers report Flock was involved in recovering half or more of the stolen vehicles in their jurisdiction. Those figures come from Flock's own impact reporting and should be read as company claims rather than independent findings. They are also not nothing. Somebody's grandmother with dementia gets found. Somebody's stolen work truck comes back. Saginaw Township credited the system with identifying two people hitting local stores within three months of installation.

Flock's trust pages state the boundaries clearly: no facial recognition, no biometric data, no predictive policing, no demographic profiling, no risk scores, searches based on observed vehicle characteristics, and local agencies deciding placement, access, and rules.

On August 13, 2026, Langley announced a package of platform changes making previously optional guardrails mandatory. Default retention drops from 30 days to seven, with exceptions for data tied to a case number and with agencies free to set their own periods. The company's justification is its own usage data showing officers finish more than 90 percent of searches without a full plate within a week. All law enforcement customers must activate Audit Assistance, which analyzes user activity and flags unusual searches for administrator review, with proactive lockouts that automatically suspend a user meeting abnormal behavior criteria pending review. More than a third of customers had already adopted it, and Flock says reviews under it were associated with arrests of several officers accused of misuse. St. Charles County is the clearest documented instance: a briefing on August 12, 2026, a flagged search pattern on August 14, and a resignation shortly after. Officers must enter a case code from their records management system for every search, with emergency searches automatically flagged.

Langley also called on state lawmakers to go further than his own company can, asking them to mandate real auditing on every law enforcement data system and raise penalties for misuse. A vendor asking to be regulated more tightly than the law requires is rare enough to note. He had previously called the DeFlock movement "terrorists," a characterization he apologized for in July 2026.

Critics were not persuaded. The ACLU called the shortened retention window a possible step in the right direction and characterized the rest as retreads of measures already proven inadequate. EFF titled its response "Too Little, Too Late," noting the shift in tone came only after a year of documented abuses, canceled contracts, and congressional attention.

The strongest version of the criticism is not that the reforms are bad. It is that they are company policy, adopted voluntarily, revisable at any time, and enforced by nobody outside the company. Which is exactly the point Hawley made in his letter, and exactly what House Bills 5492 and 5493 and Senate Bill 1131 would change in Michigan.

10. Five things that generalize past cameras

I did not write this to tell anybody how to vote on a camera contract. Reasonable people in Saginaw Township and Bay City looked at the same technology and reached opposite conclusions, and both are defensible. What I do want to argue is that five things in this story generalize far past license plate cameras, and any organization about to buy AI software should understand them.

The pitch is not the product

Saginaw Township voted on stolen cars, missing persons, and homicide suspects. Waterford Township voted on plate readers in 2022 and now has drones and integrated school district video. Nobody lied. The platform kept shipping features on the vendor's roadmap rather than the buyer's. Before you sign, the question is not what the software does in the demo. It is who decides what it does next. Ask about the roadmap. Ask which capabilities exist in the platform that you are not turning on today. Ask whether a feature or a setting can be enabled without your approval. Ask what happens to your data if you leave, and read the termination clause the way Oakland County read its auto-renewal clause.

The default configuration is the real configuration

Mountain View had statewide lookup active on 29 of 30 cameras for 17 months and did not know. Ventura County disabled national lookup in June 2023 to comply with state law and found 364,000 unauthorized accesses anyway. Pflugerville told its residents it was excluded from statewide and nationwide databases. The lesson is not that vendors are dishonest. It is that nobody at any of those agencies had pulled the logs to check whether the settings they believed were in place actually were. If you deploy software with a sharing or integration setting, somebody on your team owns verifying that setting on a schedule, and that person does not work for the vendor. Ventura Police added daily verification checks. That is the correct response.

Probabilistic output is not fact, and neither is the database behind it

Every wrongful stop above traces to the same root. A model produced a guess with a confidence score attached, and a human treated it as certainty because it arrived on a screen looking authoritative. The Aurora officers could see the vehicle in front of them was an SUV and not a motorcycle. Nobody had built a verification step, so nobody took one. And Los Angeles adds the second half of the lesson: the model can be right and the answer still wrong, because the list it checked against was stale. If your team runs AI output into a decision with consequences, the verification step has to be a designed part of the workflow, and your data hygiene has to be audited on the same schedule as your model.

Aggregation changes what the data is

One photograph of a truck at an intersection on a public road is not private in any meaningful sense. Ten thousand timestamped photographs of the same truck is a map of a person's life: where they worship, who they visit, which doctor they see, which meeting they attend on Tuesday nights. That transformation, from individually trivial records into a collectively revealing profile, is the reasoning the Supreme Court applied in Carpenter and again in Chatrie. It is also the thing most people miss when they say there is no expectation of privacy in public. The same logic applies to the operational data your own company is quietly accumulating about employees and customers.

Rules written around the wrong noun do nothing

A city ordinance banning facial recognition does not touch a system that finds people by typing "male with tattoos in a grey shirt." Description search reaches the same outcome through a mechanism the rule never named. If your AI policy prohibits specific named technologies rather than specific outcomes and specific decisions, it will be obsolete before the ink dries. Write policy around what you will not allow to happen, not around the brand of software that happens to do it today.

None of that requires anybody to decide whether Flock is good or bad. It requires understanding the technology well enough to ask the right questions before the vote, before the signature, before the rollout.

11. Common questions

What is Flock Safety and what do its cameras do?

Flock Safety is an Atlanta company founded in 2017 that sells automated license plate reader cameras and related public safety software. A fixed Flock camera photographs each passing vehicle, reads the plate with optical character recognition, generates a description of the car itself including color, body style, roof racks, and bumper stickers, and logs the time and location. The records upload to Flock servers where police can search them. The company reports roughly 20 billion vehicle scans per month.

Do Flock cameras use facial recognition?

No, and Flock states this plainly on its trust pages. The company says its system captures vehicle characteristics rather than identity, uses no facial recognition, no predictive policing, and no demographic profiling. That statement is accurate and also incomplete. Flock's Condor video cameras run a separate feature called FreeForm that accepts a plain-English description of a person and returns matching frames. No biometric template is created, which means no rule written around the phrase "facial recognition" applies to it.

How long does Flock keep license plate data?

The long-standing default was 30 days. On August 13, 2026 CEO Garrett Langley announced the default would drop to seven days, with exceptions for data tied to a case number, and with agencies still free to set their own periods. Because retention is a customer setting rather than a legal limit in most states, the real answer depends on your local agency's configuration. Michigan's pending bills would cap it at 14 days by statute.

Can out-of-state or federal agencies search my local cameras?

It depends on the sharing tier your local agency selected, and multiple communities have found their setting was not what they believed. Flock told Congress roughly 75 percent of law enforcement customers opted into National Lookup. Pflugerville, Texas found 459 outside organizations had run about 1.6 million searches touching its network. Mountain View, California found statewide lookup active on 29 of 30 cameras without its police department's knowledge. Flock says federal organizations were removed from statewide and national lookup in August 2025 and limited to one-to-one relationships.

How accurate are the cameras?

Three numbers, measuring three things. Flock markets more than 96 percent character-level accuracy and has stated 93 correct reads per 100 plates in litigation. Roseville, California reviewed 1,427 of its own alerts across 2023 and 2024 and found 71 percent contained misread plates. The Los Angeles Police Department Inspector General found 161 of about 498 stolen-vehicle alerts wrong in a July 10, 2026 audit, but traced most of those errors to stale police hot list entries rather than camera misreads. Model accuracy and alert accuracy are different problems.

How many Flock cameras are in Michigan?

More than 180 Michigan law enforcement agencies use Flock, close to a third of all agencies in the state, up from roughly 60 in 2022. Bridge Michigan counted at least 125 county or municipal agencies under contract with cameras in 50 of Michigan's 83 counties. The crowdsourced DeFlock map lists about 3,557 reported reader locations statewide, ranking Michigan eighth nationally. DeFlock entries are volunteer-contributed and are not proof of an active contract.

Is there a Michigan law regulating license plate readers?

Not as of September 9, 2026. Three bills are pending. House Bills 5492 and 5493, introduced January 29, 2026 by Representatives Doug Wozniak and Jimmie Wilson Jr., are tie-barred and sit in House Judiciary with a House Fiscal Agency analysis dated April 22, 2026. Senate Bill 1131, from Senator Jim Runestad, was introduced in August 2026. All three would cap retention at 14 days and require a warrant or court order to go further. Nationally, 21 states have an ALPR statute on the books and 28 set no retention limit in law at all. Michigan is in the second group.

How do I find out if my plate has been searched?

HaveIBeenFlocked.com aggregates Flock audit logs released through public records requests and agency transparency portals. As of August 16, 2026 the site reported 4,684,253 plates appearing across 241,961,192 logged searches. A result shows that someone ran your plate and the reason they typed, not that your car passed a camera. Absence from the database means the relevant logs have not been released, not that you were never scanned.

Can my city get the audit log showing who searched its cameras?

Sometimes. Pflugerville, Texas obtained about six months of audit records through a public information request. Genesee County, Michigan denied a similar request on the grounds that it did not own the system or the data, and is being sued in Genesee County Circuit Court by Kettering University student Josh Tewolde with the Mackinac Center Legal Foundation. Kent, Washington determined the network audit was not a public record at all because Flock generates and holds it. That question is unsettled in Michigan.

Which Michigan communities have canceled or rejected Flock?

Bay City, Ferndale, Westland, Clawson, Milford, Rochester, Sumpter Township, Menominee, Oceana County, and Dickinson County have canceled, declined to renew, or voted down agreements. The Lansing City Council passed a non-binding resolution 7 to 1 in August 2026 asking the administration and police to end the city's contracts. The list keeps growing, so verify before citing it.

Are Flock cameras constitutional?

Courts have mostly said yes for fixed cameras in public places, including appellate and federal district courts in at least 14 states plus the Ninth and Eleventh Circuits. On January 27, 2026 Chief Judge Mark S. Davis ruled for the City of Norfolk in Schmidt v. Norfolk, writing that "at least in Norfolk, Virginia, the answer is: not today." That case is on appeal to the Fourth Circuit. The Supreme Court's June 29, 2026 decision in Chatrie v. United States, holding 6 to 3 that obtaining geofence location data is a Fourth Amendment search, may change the analysis.

What should a township board or city council ask before voting?

Which sharing tier will be enabled, and who verifies it on a schedule. Whether the contract contains the perpetual license to disclose agency data, and what the February 2026 terms changes mean for termination. Which platform capabilities exist that you are not turning on today, and whether the vendor can enable any of them without your approval. Whether any pilot converts to a paid contract automatically, the way Oakland County's nearly did. Who reads the audit log, how often, and what triggers a review. What happens to collected data if you terminate.

12. How to check your own community

Here is the simplest thing in this entire article. Sam Aly in Pflugerville filed a records request. A council member in Ferndale read his own city's logs. Somebody in East Lansing asked for 1,700 pages. Josh Tewolde in Flint went to court when Genesee County said no.

In every case, the document that changed the outcome already existed. Nobody had asked for it.

Four steps, none of which require a lawyer:

  1. Search your plate. HaveIBeenFlocked.com covers audit logs that have already been released. It is free and it takes 30 seconds.
  2. Map your area. DeFlock maintains a crowdsourced map of reported reader locations. Treat the operator labels as volunteer metadata, not proof of contract.
  3. Check for a transparency portal. Some departments publish one voluntarily. Saginaw Township launched its portal on August 24, 2026 at transparency.flocksafety.com. A portal shows how the local department uses the system. It does not show who outside the community is searching it.
  4. File the request that matters. Ask your city, township, or county for two documents: the full Flock contract, and the Flock network audit covering a specific date range. The network audit is the one that lists outside organizations and their search counts. Search the contract for the phrase "worldwide, perpetual, royalty-free" and for the auto-renewal terms. If the response says the audit belongs to the vendor, that is the same answer Genesee County gave, and it is being litigated.

Michigan has more than 180 agencies under contract, roughly 3,557 mapped cameras, and no state law. Most of those communities have never pulled their own audit. The Pflugerville ratio was 315 outside searches for every local one. The Ferndale ratio was more than 99 to 1. Nobody knows what the ratio is in Saginaw Township, or Flint, or Waterford, or Lansing, because nobody has looked.

That is a solvable problem, and it costs the price of a records request.

Sources and method

Reporting drawn on for this article includes 404 Media, WIRED, The Washington Post, the Associated Press, The Texas Tribune, the Austin American-Statesman, CBS Austin, Community Impact, KQED, CBS Los Angeles, the Mountain View Voice, the San José Spotlight, the Wisconsin Examiner, East Lansing Info, Bridge Michigan, the Detroit News, Michigan Advance, Michigan Public, WILX, WNEM, WLNS, WKAR, WOODTV, Michigan Capitol Confidential, GovTech, Police1, Courthouse News Service, The Record, Cybernews, Newsweek, Fortune, and Reason.

Public records and analysis: the Electronic Frontier Foundation, the Institute for Justice and its ALPR abuse and cancellation databases, the ACLU and the ACLU of Massachusetts Data for Justice Project, the University of Washington Center for Human Rights report "Leaving the Door Wide Open" (October 21, 2025), the Illinois Secretary of State audit findings (August 25, 2025), the Los Angeles Police Department Office of Inspector General audit (July 10, 2026), the Georgia Bureau of Investigation, Senator Josh Hawley's letter to Garrett Langley (August 26, 2026), the Wyden and Krishnamoorthi letter to the FTC (November 3, 2025), Michigan House Fiscal Agency analysis of House Bills 5492 and 5493, the Mackinac Center Legal Foundation filing in Tewolde v. Genesee County, and Flock Safety's own trust pages, network sharing statement, product documentation, blog posts, and public statements.

Where sources conflicted on a figure, both are given with attribution. Where a claim could not be independently verified, it was cut rather than hedged. Company impact statistics are labeled as company claims throughout. Legislative status, contract counts, and litigation posture in this area change frequently. Verified as of September 9, 2026.

Tim Bish is the founder of Understanding Your AI, which provides AI education, training, and consulting for businesses and nonprofits in Michigan's Great Lakes Bay Region.

TB

Tim Bish

Tim Bish is a former school principal and system administrator who teaches Michigan teams to use AI through Understanding Your AI. He also used AI to run marketing and training solo across five business units and a nonprofit at Vantage Plastics. More about Tim · See workshops

Buying AI software this year?

The five lessons in this article are the ones I walk boards and leadership teams through before they sign anything. If your organization is evaluating an AI vendor, a free assessment is a good place to start.

One AI thing you can use on Monday.

A weekly read for people with actual jobs. The week's signal, the two videos worth your time, and one prompt that works. No hype. Free.