This is the third piece in CDT’s “AI In Policing” series, in which we explore how police use various AI technologies for surveillance and investigations, the risks posed, and how we should respond.
Automatic License Plate Readers (ALPRs) have tracked American drivers for over a decade. But the recent integration of sophisticated AI and the explosive growth of private-sector networks, most notably Flock Safety, have amplified the risks of ALPRs even further. These surveillance systems threaten to become a near-ubiquitous, permanent, and searchable record of daily life for tens of millions of law-abiding Americans who rely on personal vehicles every day. This blog examines how ALPRs claim to improve public safety, the grave risks they pose to privacy and civil liberties, and how policymakers must intervene before dragnet surveillance becomes the American norm.
ALPRs are surveillance systems that permit effortless, automated tracking of vehicles across expansive areas. ALPRs use cameras that are mounted on streetlights, overpasses, and police cruisers — and sometimes covertly behind disguises like traffic cones and sand barrels. These high-speed cameras capture images of every passing vehicle and license plate, and then process those images using optical character recognition to “read” the alphanumeric characters on the license plate and convert them into machine-readable text.
ALPRs are used for surveillance in two primary ways: Sometimes these systems automatically check license plates against databases or “hot lists” for stolen vehicles, missing persons, or wanted persons, which can trigger automatic alerts to law enforcement. These alerts can be used to dispatch an officer to make an arrest or find a missing person. Other times the data is merely stored — sometimes indefinitely — for later manual search or review, such as to map out a vehicle’s movements over the past several months. In many cases, the data is integrated into national networks of ALPR databases that may be accessible to state and federal law enforcement entities beyond the jurisdiction where the data was collected. This means the government can map out a vehicle’s movements not only for extended periods of time, but also across huge swaths of the country.
ALPRs have been around for decades, but modern ALPRs are increasingly using artificial intelligence (AI) to collect and produce far more sensitive data that can be used to surveil everyone who gets behind the wheel. Companies like Flock Safety, Motorola Solutions, and Axon all now offer some form of “vehicle fingerprinting,” which uses computer vision models to identify the make, model, color, and unique attributes of a car, such as a roof rack, custom bumper sticker, or a dent. This has enabled police to search for “motorcylce [sic] with hells angels,” and “bmw with cracked sunroof,” for example. Many companies also offer some form of “convoy analysis,” which purports to identify vehicles traveling together. This could reveal sensitive personal associations, such as who may have been traveling together to attend a political rally in another state.
The massive datasets generated by these cameras are now also being combined with other public records and queried using natural language interfaces. Instead of searching for a specific license plate, law enforcement can use natural language queries that allow them to search through billions of records to find more nuanced details and patterns of behavior. This makes it possible to conduct broad, speculative searches, such as to identify all vehicles that drove by a particular bank at least twice in the days prior to a bank robbery.
Many ALPR systems are also beginning to offer automatic anomaly detection or other predictive analytics. This allows AI systems to identify “unusual” patterns or behaviors that might indicate security concerns before law enforcement are aware of a crime, or even before any crime has occurred.
KNOWN USES
Map shared with permission from DeFlock.
ALPRs are basically ubiquitous. Motorola Solutions claims to have cameras from more than 3,700 agencies and business partners in its ALPR data-sharing network, and Flock Safety claims to be in over 6,000 communities, with its cameras capturing over a billion images per month. Federal law enforcement agencies, like ICE, FBI, and CBP, also spend tens of millions of dollars on ALPR systems and data.
But a growing wave of cities — including Denver (CO), Austin (TX), and Cambridge (MA) — have terminated or declined to renew their contracts in recent months, especially with Flock. Some have even gone so far as to cover the cameras with trash bags to ensure Flock does not still have access to the video feeds. Much of this backlash centers on unauthorized side-door access into these systems for federal immigration enforcement. Despite Flock branding itself as a local tool, investigative reports throughout 2025 revealed that federal agencies like ICE and CBP have repeatedly accessed local Flock data, often through undisclosed “pilot programs” or by having local officers run searches on their behalf. In cities like Berkeley and Portland, activists and policymakers have sounded the alarm that these cameras are being weaponized to bypass local regulations prohibiting assistance with immigration enforcement and can turn taxpayer-funded city infrastructure into a deportation dragnet.
RISKS
How it might not work…
Despite marketing claims, ALPRs are far from infallible, and mistakes can have immediate and dangerous real-world consequences.
- Misidentification: ALPR systems may mistake a “2” for a “7” or an “H” for an “M,” especially in conditions like rain, snow, and mud. Even such a simple technical glitch can lead to a dangerous real-world confrontation. These “false hits” can trigger law enforcement to effectuate felony stops, which are high-risk procedures where officers may approach a vehicle with weapons drawn, believing the driver is a dangerous or wanted criminal. In one case, two innocent individuals were detained at gunpoint in front of their family, only for police to realize minutes later that the AI had misidentified a character on their license plate.
- Stale data: ALPR systems often rely on “hot lists” — databases of stolen cars, missing persons, or wanted persons — to generate real-time alerts for law enforcement. If these lists are not updated in real-time, previously stolen cars that have already been recovered or individuals that are no longer “wanted” may be repeatedly pulled over and detained based on stale data. This can lead to tense and potentially dangerous situations, like what happened when two brothers were stopped and detained at gunpoint because the authorities failed to remove their vehicle from the stolen list after it had been recovered.
Even if it does work…
The most profound dangers of ALPRs may arise when the technology works exactly as intended. A “perfect” ALPR system creates a comprehensive map of movement.
- Pervasive tracking: Unlike human officers, ALPRs never get tired and never stop watching. By documenting the location of a vehicle at specific times, police can reconstruct elements of a person’s entire life: where they live and work, with whom they associate, and which sensitive locations they visit (e.g., doctors’ offices and places of worship). Such use of ALPRs for pervasive tracking raises concerns very similar to cellphone location tracking that the Supreme Court ruled requires a warrant in 2017.
- Abuse and stalking: Because many ALPR systems lack robust access controls and meaningful oversight, they can be easily transformed into personal harassment tools. When an officer can type a plate number into a search bar and instantly reconstruct a person’s movements over weeks or months, harassment and stalking is easy. Officers have misused these systems to monitor ex-partners, track ordinary citizens, follow their personal romantic interests, and surveil political protestors. In one case, an officer used an ALPR to track his ex-girlfriend over 200 times. There is also evidence that ALPRs have been used to track someone seeking reproductive healthcare and are another instrument for racist policing.
- Predictive stops: ALPR systems can be used to analyze driving patterns and flag vehicles whose movements an algorithm deems suspicious even in the absence of any specific, explainable evidence indicating criminal activity. Police may then initiate traffic stops pretextually based on minor infractions, such as a cracked windshield or hanging air freshener. In fact, the U.S. Border Patrol monitors millions of American drivers nationwide in this exact way, identifying and detaining people whose travel patterns it deems suspicious. This infringes Americans’ freedom of movement, and could happen when a vehicle is tagged at “anomalous” locations or times, even when no evidence suggests that the particular vehicle is engaged in any specific criminal activity.
- Private-public dragnet: Companies like Flock Safety and Motorola Solutions have created a privatized surveillance network. By encouraging neighborhoods, businesses, and schools (and even school buses) to install cameras that feed directly into police databases, they are creating a seamless web of surveillance that often bypasses traditional public oversight and procurement transparency. As one mayor noted after his city discontinued their Flock contract but the company left their cameras in place, “The fact that they didn’t take the cameras down shows that we are the product.”
RECOMMENDATIONS
To protect civil liberties from automated tracking, policymakers should adopt the following:
- Warrant requirements for criminal investigations: Most federal and state courts that have considered the issue have determined that querying ALPR systems does not require a warrant under the Fourth Amendment, which highlights the importance of establishing statutory warrant protections. Montana and Minnesota have already established statutory warrant protections for law enforcement queries in some circumstances and offer useful starting points.
- Warrants for extensive historical location information: Querying an ALPR database to track a person’s movements over time for use in criminal investigations should require a search warrant based on probable cause.
- Warrants to identify vehicles near the scene of a crime: Querying an ALPR database for records of vehicles that were near the scene of a crime when it occurred should require a warrant that features strict limits on geographic scope and timeframe to ensure searches are narrowly tailored to the investigations, and to minimize the impact on people unlikely to be involved.
- Warrants for hot lists that generate real-time alerts: Hot lists that generate real-time alerts should only generate alerts for vehicles that are owned or operated by someone subject to an active arrest warrant, or vehicles that are — pursuant to a warrant supported by probable cause — determined to be involved in the commission of a felony.
- Exceptions: These warrant requirements should be subject to commonsense and non-controversial exceptions, such as for emergencies, stolen vehicles, and locating missing persons.
- Strict retention limits: Agencies should adopt policies that require the timely deletion of ALPR data that does not generate a hit against specific hot lists or is not relevant to a pending criminal investigation or civil matters like speeding infractions or tolling. Storing the movements of innocent drivers for months or years “just in case” creates a massive repository ripe for abuse.
- Restricted access and data sharing: Access to ALPR data should be strictly limited to personnel who need it for their job functions. Governments should also adopt rules and appropriate punishments for unauthorized access or misuse of ALPR data to address the abuses discussed above like stalking and surveillance of protected free speech. Furthermore, Government entities should only provide access to vehicle data collected or produced by ALPR systems to out-of-state law enforcement agencies if the requesting entity is also subject to recommended privacy protections, such as warrants. They should also require ALPR providers with which they contract to agree to appropriate privacy and use restrictions, such as a prohibition on the sale of ALPR data to data brokers or other commercial third parties.
- Mandatory transparency and auditing: Every search of an ALPR database must be auditable and require a logged justification, including a specific case number and a declared offense type where appropriate, to prevent officers from using the system for personal or speculative reasons and to detect unauthorized use. These logs must be subject to regular, independent audits to detect patterns of misuse or disparate impacts on specific neighborhoods and populations. Furthermore, agencies should be required to publish annual transparency reports that disclose the number of cameras deployed, the number of hits generated, and how often those hits actually led to an outcome such as a conviction, locating a missing vehicle or person, enforcement of a traffic citation or other public benefit so the public can weigh the actual efficacy against privacy costs.
- Discovery and preservation rights: Rules for criminal legal proceedings should ensure that defense attorneys can access the full audit trail of any ALPR search used against their client, including original AI-generated hits and subsequent human verifications. To prevent the loss of potentially exculpatory evidence, such as when a defendant claims ALPR data will prove they were not at the scene of a crime, laws should require agencies to honor reasonable preservation requests from defendants for data related to their own location if that data is relevant and material to their defense.
CONCLUSION
While ALPRs have largely flown under the public radar until recently, they represent one of the most significant expansions of surveillance power in decades. For a large portion of the American public, cars are a necessity of life. By automating the tracking of all vehicles across huge portions of the country, we risk creating a society where freedom of movement is a relic of the past. Without policy intervention, these tools will continue to build a permanent record of our lives, one plate at a time.