Automated license plate readers were sold as tools for finding stolen cars and wanted suspects. As those cameras become part of nationwide, AI-searchable networks, the harder question is what government should be allowed to learn from the millions of vehicles they photograph.
For years, the debate over automated license plate readers could be reduced to a relatively simple question: Should police be allowed to photograph a license plate that is already visible on a public road?
That question is becoming obsolete.
The cameras still read license plates. They still alert officers when a stolen vehicle, wanted suspect or missing person passes by. Police departments across the country can point to real cases in which that capability helped officers find people faster and solve crimes that might otherwise have gone cold.
But the technology surrounding those cameras has changed substantially.
What was once essentially a digital license plate reader has become part of a much larger system capable of storing vehicle movements, sharing information across jurisdictions, identifying vehicle characteristics and searching enormous collections of historical data.
Now, artificial intelligence is beginning to sit on top of that network.
WIRED reported Wednesday that Flock Safety, one of the country’s largest providers of automated license plate reader technology, is testing a new artificial-intelligence system called OS Investigate. The system, according to WIRED’s analysis of code hosted by Flock, can combine plate-reader information with police records, dispatch logs and commercial identity databases and allow investigators to begin some searches without knowing a person’s name or license plate at all.
Flock says the product remains in development, is being tested by a small number of law enforcement partners and is separate from its traditional license plate reader product. The company also said capabilities could change before any wider release.
That distinction matters.
So does the direction of travel.
The central public-policy question surrounding automated license plate readers is increasingly not whether a police officer should be able to identify a stolen Chevrolet passing an intersection.
It is what happens after millions of ordinary vehicles are photographed, indexed and connected.
From Camera to Network
A traditional automated license plate reader performs a relatively intuitive function. A camera photographs a vehicle, software reads the plate and police can compare that plate with a list of vehicles associated with crimes, warrants or missing persons.
Modern systems collect more.
Killeen’s own description of its Flock system, for example, says its cameras can identify not only license plates but attributes including vehicle type, color, make and accessories. The city’s 2025 agreement authorized 67 cameras, five of which were described as having “real-time surveillance capability,” integrated with the Killeen Police Department’s Real Time Crime Center.
The larger advantage comes from scale.
Flock operates a network used by thousands of law enforcement agencies around the country. Agencies can share access across jurisdictional lines, meaning a vehicle photographed in one city may potentially help investigators working a case somewhere else. In Texas alone, more than 200 local law enforcement agencies have data-sharing agreements with the Texas Department of Public Safety involving the technology.
From an investigative perspective, the appeal is obvious.
A suspect does not stop existing when he crosses the city limits.
A stolen vehicle from Temple may turn up in Dallas. A homicide suspect wanted in another state may drive through Central Texas. A missing child may be hundreds of miles from where the initial report was filed.
A network can find things an isolated camera cannot.
The same characteristic that makes the technology useful, however, creates its central civil-liberties problem.
The network does not photograph only suspects.
It photographs everyone who drives past.
When the System Gets It Wrong
The privacy debate assumes the system is identifying the right vehicle. Sometimes it does not.
A recent Los Angeles Police Department inspector general review found that Flock cameras scanned more than 210 million plates during a two-month period and generated 161 stolen-vehicle alerts that ultimately proved incorrect. The same review found 337 alerts that did lead to the recovery of stolen vehicles — a useful reminder that the technology can simultaneously work and make mistakes.
Sometimes the camera is not even the source of the error. In Plymouth, Minnesota, police stopped a driver after Flock cameras twice flagged his plate as stolen. Officers later discovered that another agency had entered an incomplete plate number into the national stolen-vehicle database. Flock correctly matched the information it had been given; the information itself was wrong.
That distinction may matter very little to the innocent driver surrounded by police officers.
Automated systems create a temptation to treat a computer-generated alert as objective truth. Flock itself says an alert should be one part of an investigation rather than the sole basis for a stop.
The question is therefore not simply how accurate the cameras are. It is how much human verification should occur before an automated observation becomes government action against a person.
Flock Is Already Changing Its Rules
The debate has intensified nationally as examples of misuse and concerns over cross-jurisdictional access have accumulated.
Part of the criticism has focused not merely on what Flock’s systems are capable of doing, but on whether the safeguards governing those capabilities have actually worked.
Some existing controls have relied heavily on users policing themselves. One Flock safeguard, for example, required officers to enter a justification before conducting certain searches. Public records reviewed by civil-liberties advocates found that officers could satisfy the requirement with little meaningful oversight. In one Oregon department, records showed 111 searches justified simply as “investigation” and another 20 entered with the word “hehehe.”
The problem was not that the system lacked a rule. The problem was that typing something into a box was effectively being treated as compliance.
That distinction matters when the database being searched contains information about the movements of people who are not suspected of crimes.
Last week, Flock announced a series of changes to its platform intended to address those concerns. Beginning Jan. 1, the company says law enforcement customers will be required to use tools designed to identify abnormal search behavior and tie searches to case numbers. Flock also plans to reduce its standard data-retention period from 30 days to seven, although information connected to an investigation can be retained longer.
The changes are significant because some of those protections were previously optional.
Flock says the new controls will increase accountability while preserving the investigative value of the system. Critics argue that the fundamental problem remains: access to large collections of location information can still occur without the judicial oversight normally associated with more traditional searches.
More than 50 jurisdictions have ended relationships with the company amid the broader debate, according to the Associated Press. At the same time, police agencies continue to defend the technology as an important tool for recovering stolen vehicles, locating missing people and finding suspects in serious crimes.
Both of those things can be true.
A technology can solve crimes.
It can also deserve rules.
In fact, the more powerful and useful a surveillance technology becomes, the more important those rules may be.
Then Comes Artificial Intelligence
That brings the debate to OS Investigate.
The system includes prewritten prompts allowing investigators to query patterns across datasets. Some searches described in the code could begin with a place and time rather than a known suspect — for example, identifying vehicles repeatedly appearing within a particular neighborhood or finding vehicles that traveled through multiple locations in a particular sequence.
Other tools described could connect vehicle information with names, addresses, relatives, arrest information, police case files and other records. The system can also identify vehicles that repeatedly appear together, potentially allowing investigators to infer associations among drivers.
There are important limitations to what is currently known.
This information was reconstructed with portions of the product from code Flock’s website delivered through its login pages. The publication did not have access to Flock’s backend systems and could not determine exactly how every request would function once submitted. Flock said the product remains under development and that the eventual capabilities and workflow may change substantially.
That means OS Investigate should not be treated as though every Flock customer currently possesses those capabilities.
There is also no evidence available to The Directory at publication time establishing that Bell County law enforcement agencies are using OS Investigate.
But its development illustrates how quickly the debate has moved.
The question is no longer simply:
Did my car pass a camera?
Increasingly, it is:
What can a computer learn when that photograph is combined with thousands of other photographs and other databases?
The Difference Between Observation and Reconstruction
Americans have never had an absolute right to anonymity while driving down a public highway.
Police officers can watch traffic. They can see a license plate. They can follow a vehicle under appropriate circumstances. Businesses and homeowners routinely operate security cameras that capture portions of public roads.
Technology changes the scale of that observation.
One officer standing beside a road can observe one place at one moment.
A sufficiently dense network can potentially remember many places over many days.
That distinction is at the center of emerging legal arguments surrounding automated plate readers. Courts have generally been reluctant to treat isolated photographs of vehicles traveling on public roads as Fourth Amendment searches. But judges have also acknowledged that increasing camera density and technological capability could eventually create something fundamentally different from an officer simply observing a car in public.
The unresolved question is when individual observations become a reconstruction of someone’s life.
A trip to a grocery store reveals very little.
Repeated trips to an oncologist may reveal more.
So might visits to a church, mosque, political headquarters, union hall, gun store, addiction-treatment center or romantic partner’s house.
None of those locations is necessarily private.
The pattern can be.
That is why retention periods, search requirements, auditing, outside-agency access and judicial oversight matter at least as much as the camera itself.
What Does the Fourth Amendment Protect?
There is a strong argument in favor of license plate readers that civil-liberties advocates sometimes glide past: Americans generally do not have a reasonable expectation that their movements on public roads will remain unseen.
The Supreme Court said as much in United States v. Knotts in 1983, holding that a person traveling on public roads has no reasonable expectation of privacy in movements visible from one place to another. The Court reiterated that principle decades later.
A police officer does not need a warrant to see a car drive past him. A license plate is intentionally displayed in public. Cameras have existed in public spaces for decades.
If that were all Flock did, the constitutional argument would be considerably simpler.
But Supreme Court doctrine has also become increasingly wary of the government’s ability to use technology to accomplish surveillance at a scale that would once have been practically impossible.
In United States v. Jones in 2012, the Court held that placing a GPS tracker on a vehicle and monitoring it constituted a search, although the majority’s decision relied heavily on the government’s physical intrusion onto the vehicle. Justices Sonia Sotomayor and Samuel Alito separately raised a broader concern: prolonged electronic monitoring can reveal a comprehensive picture of someone’s public movements even when each individual movement could have been observed legally.
Then came Carpenter v. United States in 2018. The Court held that obtaining an extended history of a person’s cell-site location information generally constituted a Fourth Amendment search. The Court deliberately described its decision as narrow and did not condemn ordinary security cameras, but it recognized something important about the digital age: location information collected over time can reveal far more than any individual observation.
That leaves modern ALPR networks sitting in an unresolved space between old principles and new technology — exactly the tension already identified in the underlying research.
Seeing my car at Main Street and Central Avenue at 3 p.m. is one thing.
Building a searchable database showing where my car has appeared for days or weeks is something else.
And this is where I think government should bear the burden of answering a basic question:
Just because information can legally be observed in public, does that mean government should be entitled to collect, store, aggregate and search it indefinitely without judicial oversight?
The Fourth Amendment was written long before license plate cameras, cloud databases or artificial intelligence. Its principle was not.
Government power requires limits.
The constitutional question confronting courts now is where those limits belong when surveillance that once would have required dozens of officers following someone around the clock can increasingly be accomplished by typing a query into a computer.
This Is Already a Bell County Question
This debate is not confined to Washington, Silicon Valley or major metropolitan police departments.
Bell County law enforcement agencies already use automated license plate readers, including Flock systems.
Killeen approved its nearly $488,000 two-year Flock agreement in March 2025. The system includes 67 cameras integrated into the department’s Real Time Crime Center.
Temple uses a combination of Flock and other ALPR platforms throughout the city and on police vehicles. The department’s publicly posted policy says the system is intended to detect license plates and vehicles rather than facial characteristics, and lists immigration enforcement, harassment, intimidation, personal use and use based solely on protected characteristics among prohibited uses. Temple also says system access requires a valid reason and that hot-list alerts must be verified by a person before police action is taken.
Temple has also published numerous examples in which ALPR information produced tangible investigative results.
This year alone, the department says Flock helped officers locate suspects wanted in shootings, thefts and homicide investigations; identify vehicles involved in hit-and-run crashes; recover stolen vehicles; and assist agencies elsewhere in Texas and in other states. In one March case, Temple police used Flock information to help locate a homicide suspect wanted in Washington state, who was later arrested with assistance from Killeen and Harker Heights police.
Those cases matter.
A serious examination of surveillance technology should not pretend it has no benefit merely because the privacy questions are uncomfortable.
But success stories answer only one part of the public-policy equation.
Residents are also entitled to know how often the systems are searched, who can access the information, how long it is retained, which outside agencies receive access, what safeguards prevent misuse and whether newer analytical tools are available locally.
We do not yet have complete answers to those questions.
LOCAL REPORTING IN PROGRESS
The Directory has requested additional information and comment from Bell County-area law enforcement agencies regarding their use of automated license plate reader technology.
The questions include current camera inventories, data-retention practices, search and audit procedures, interagency data sharing, access by state or federal agencies, documented misuse, and whether departments have access to newer Flock analytical products or capabilities beyond traditional license plate searches.
Public Information Requests are pending, and agencies have been given an opportunity to respond.
This article will be followed by a full Bell County analysis once those records and responses are available.
The Wrong Question
Privacy debates often collapse into two positions that are equally unsatisfying.
One says that if someone has done nothing wrong, there is nothing to fear from government collecting the information.
The other treats virtually any new law-enforcement technology as inherently illegitimate.
Neither does much to help communities govern technology that is already here.
The useful question is not whether automated license plate readers are good or bad.
They are tools.
The questions are what they can do, who can use them, under what circumstances, for how long and with what oversight.
Flock’s own changes suggest the company recognizes that some rules surrounding its network needed strengthening. Its simultaneous development of increasingly sophisticated AI investigative tools suggests the capabilities of that network are not standing still.
That creates a problem familiar throughout the history of technology and government:
Policy moves slowly.
Software does not.
City councils may believe they are approving cameras to help police find stolen cars. Several software updates later, the same physical infrastructure may be capable of something considerably more sophisticated.
That does not mean every new capability should be prohibited.
It does mean the public deserves another opportunity to decide where the boundaries belong.
Because the hardest question surrounding modern surveillance is no longer whether the government can see you driving down a public street.
Of course it can.
The question is how much the government should be allowed to remember about where everyone has been — and what it should be permitted to discover once machines can connect those memories together.