Police Are Pulling Drivers Over Because AI Calls Their Travel "Suspicious", the ACLU Warns

On October 26, 2026, the ACLU published an essay by Jay Stanley with a blunt claim: police are increasingly pulling people over because AI deems their driving patterns "suspicious". The mechanism is simple to describe. Cameras log license plates everywhere, software scores where a car has been, and an officer acts on the score with a traffic stop that cites something minor. For readers who follow AI policy, it is a concrete case of a model output turning into a physical, legal encounter. This post separates what the essay documents from what it argues, and links the earlier explainx.ai coverage of the same surveillance stack.

TL;DR: what is claimed and what is documented

Agents that act on your behalf need rules. Agentbeam stops the dangerous actions →

The account at the center of the essay

Stanley opens with Grace Knowles, a young woman from northern Virginia who took a road trip through Oregon, Texas and Louisiana in September 2025. On her way to Florida, the essay says, an officer pulled her over and told her she had been in the left lane too long. He then asked where she had been and for proof she had come from New Orleans. She showed a pamphlet from a whiskey museum.

According to Knowles, the officer then said he was "part of the FBI" and that when he scanned her plate he "could see you had suspicious travel patterns". She told Stanley he doubted her story because "he didn't show any hits" in New Orleans. He invited her into his car and showed her software that looked like a heat map of the United States, with circles where she had driven and a photo of her plate at each one.

This is a single account, reported by the ACLU from the driver's own telling. The essay does not name the software and does not claim the officer's agency confirmed any of it. Treat it as an illustration, not a measurement. Stanley himself calls it a likely mild version, saying that for many people such encounters will be "far more fraught, frightening and dangerous".

How AI turns plate data into suspicion

An automated license plate reader is a camera that photographs plates and logs the time and place. Alone, a hit tells police where one car was. The shift Stanley describes is analysis across many hits: scoring a whole travel history as normal or suspicious, and finding which vehicles tend to appear together.

He traces it through three steps:

• July 2025: Flock touted the ability to alert police to drivers whose travel patterns were "suspicious" and to track associations by analyzing which vehicles are often seen with other vehicles. Stanley wrote about it then as a leap from investigating a suspicious vehicle to generating suspicion where none existed.
• A year later: Wired obtained files describing Flock's next-generation AI system, already used by some police. Per the essay, they show open-ended AI queries over plate readings and other data, including arrest records, case files and commercial data broker databases.
• Now: Stanley says the approach is spreading, with reports from across the country of "pretextual stops" based on where someone has been.

For the business and politics around that vendor, see our posts on Flock Safety cutting 18 percent of staff amid backlash and Rep. Tlaib's bill to bar federal money for Flock-style surveillance.

The wider pattern: whisper stops, intel stops and Border Patrol

The essay says police have names for these stops, "whisper stops" and "intel stops". A pretextual stop uses a claimed minor infraction as cover for the real reason. The ACLU says it has pushed to ban them for years, and that AI plus ALPR changes the scale: police can now track everyone, all the time, and let software suggest suspects.

Stanley lists several cases, each with its own source:

1. Border Patrol. The Associated Press reported in November 2025 that the agency monitors drivers nationwide inside the US for suspicious travel patterns and works with other agencies to stop and search vehicles. In the AP's account, as summarized by KPBS, a network of cameras logs plates and an algorithm flags vehicles by origin, destination and route, and federal agents can alert local police who stop drivers for minor infractions. CBP told the AP the readers help identify threats under a "stringent, multi-layered policy framework". A furniture-mover who drove from South Carolina was stopped near the border and arrested on suspicion of money laundering over cash he carried, per the essay, though the search found nothing illegal.
2. A Texas jury. The essay says a jury recently sided with Schott, an oilfield-supplies seller who sued after a stop. A federal fusion center had flagged him as suspicious for driving from Houston to Carrizo Springs and back the next day. The Institute for Justice, which represented him, noted that the same county had subjected thousands of other drivers to the same routine. Dashcam footage showed the claimed lane drift was bogus, and the officer later admitted it.
3. The Markup. In March 2026 it reported on a California grandmother questioned by CBP officers at the border about how much time she spends at local casinos, apparently based on her driving patterns.
4. Wisconsin to Michigan. In April 2025 a man was targeted because he drove between those states, which police noted was "a known source State for Marijuana as it is legal there". Stanley's point: a whole neighboring state became a suspicious destination.

Why this matters beyond privacy

This is an AI story because the model's output is the stated reason for state action. Three failure modes follow from how such systems work.

Base rates. Almost everyone who drives long distances, rents a car, works near a border or visits a casino is innocent. A system that flags unusual travel will mostly flag innocent people. Stanley makes this point directly: the algorithms will "inevitably sweep in far more innocent people than actual wrongdoers".

Unfalsifiable evidence. In the Knowles account, the absence of a plate hit in New Orleans was treated as a sign she was lying. A data gap became evidence against the driver. When a score is shown to an officer as a heat map, the driver cannot easily inspect or challenge the reasoning.

Mission creep. The essay warns that AI on top of ALPR databases enables abuse based on political or protest associations and travel, not just drug or trafficking suspicion. We saw a similar concern in ICE's request for AI to monitor detainee calls and voice profiles.

It also fits a broader run of AI-and-policing stories we have covered, including an Anthropic model that sent a false homicide tip to Philadelphia police and Raven, a tool police use to geolocate a house from one interior photo. In each, a model output enters a police workflow with little public visibility into how reliable it is.

What the essay does not establish

A careful reader should note the gaps:

• No count. There is no figure for how many stops result from AI travel scoring.
• No vendor link in the lead case. The officer's software is not named, and the essay does not say it was Flock's.
• No response. The text does not include comments from Flock, the FBI or the officer's agency.
• Mixed legal picture. The AP reported that courts have generally upheld collecting plates on public roads, while some scholars see constitutional questions about large networks. The Texas verdict shows at least one jury rejecting a pretext.
• Advocacy source. The ACLU openly argues for banning license plate surveillance, so its conclusion is a position, even where its facts are sourced.

Hacker News had the essay posted on the day it ran, but no discussion had started when we checked, so there is no community testing to report yet. We will add an update section if officials or vendors respond.

What this means for drivers, builders and policymakers

Drivers. You cannot opt out of plate cameras. You can know your rights in a stop: you may ask whether you are free to go, you need not consent to a search, and you should note the officer, agency and time. Public records requests may reveal what plate data an agency holds on your car, though rules vary by state. If you believe a stop was a pretext, a local civil rights lawyer or your state ACLU affiliate can advise.

Builders. If you ship models that score people or behavior, this is the scenario to design against: a score that is opaque to the person it affects, trained on data of unknown representativeness, with a human acting on it quickly. Log reasons, expose confidence, and test false-positive rates on ordinary behavior, not just on known bad actors. Agent systems that touch real people raise the same risk class, and tools like AgentBeam from the explainx.ai team exist to stop AI agents before they take dangerous actions.

Policymakers. The live options are the ones already on the table: funding bans like the Tlaib bill, limits on data retention and sharing, warrant or reasonable-suspicion requirements before acting on pattern scores, and disclosure of when a stop was triggered by an algorithm. The Knowles account suggests drivers are not told, which makes any later challenge harder.

What to watch next

• Whether Flock or other vendors respond to the specific claim that their travel-pattern alerts are generating pretextual stops.
• Whether more court cases follow the Texas verdict, and whether disclosure of algorithmic tips becomes a requirement.
• Whether the Wired-reported open-ended AI query features spread to more departments.
• Whether Congress moves on any federal funding limits for plate readers.

Related reading

• Flock Safety to cut 18% of staff as license-plate surveillance backlash grows
• Tlaib introduces the Get the Flock Out of Our Communities Act
• Raven: the AI tool police use to find the exact house in one interior photo
• ICE wants AI to monitor detainee calls, voice profiles and camera feeds
• An Anthropic AI model sent a false homicide tip to Philadelphia police
• A Claude diary entry was reported to police: how safety review works

Primary sources: the ACLU essay by Jay Stanley and the ACLU's earlier analysis of the Border Patrol plate reader program.

Details reflect the sources as of October 26, 2026 and may change as agencies and vendors respond.