ACLU Warns AI Now Flags Road Trips as Suspicious Driving

By Daniel Rao

Grace Knowles was not speeding. She was not swerving, not drunk, not running from anything. She was, by her own account, just a young woman between jobs, driving out of New Orleans toward Florida after a road trip through Oregon, Texas and Louisiana — and that, apparently, was enough. An officer pulled her over, questioned her about where she had been, demanded proof she had really visited New Orleans, and then showed her a heat map of her own life on a screen in his car. The stop, in the end, was not about how she drove. It was about what a machine thought her driving meant.

That is the warning at the heart of a blog post published by the American Civil Liberties Union (ACLU) on 26 October 2026, and it should unsettle anyone who has ever taken the long way home. Police, the ACLU says, are increasingly using artificial intelligence and automated license plate reader (ALPR) data to flag motorists whose travel histories look “suspicious” — even when there is no evidence they have done anything wrong. The officer does not need to catch you doing something. The system has already decided you look like someone worth catching.

Before going further, honesty demands a caveat, and this article will keep returning to it: the ACLU post is a warning built around individual accounts, not a measured nationwide count. It does not establish how many drivers have been stopped this way, which departments or systems were involved in each case, or how often a flag actually leads to a stop. What it does establish — through Knowles’s account, a Texas jury verdict, and reporting by the Associated Press, WIRED and others — is that the mechanism is real, that it has already produced at least one unconstitutional stop in a court’s eyes, and that nobody can currently tell you, with confidence, how it decides you are suspicious.

Photo: Automated license plate reader cameras scanning vehicles crossing into Pensacola Beach, Florida. File photo — not the camera system in Grace Knowles’s stop, which the ACLU does not identify. Credit: Tony Webster, CC BY-SA 2.0, via Wikimedia Commons.

A road trip becomes an interrogation

Knowles, a woman in her 20s from northern Virginia, told the ACLU she had left New Orleans that morning in September 2025 when she noticed a police car behind her. “I know I was going a normal speed, but he pulled me over anyway,” she said. “And he made up some excuse — that I’d been in the left lane too long.”

What followed was less a traffic stop than a cross-examination about her holiday. The officer asked where she was coming from and where she was going. When she said New Orleans, he asked for proof. Her proof, in the gloriously ordinary way of road trips, was a pamphlet from a whiskey museum she had visited. “I’m just a girl doing a road trip,” she told him, handing it over. “Here’s my proof.”

Only then, she says, did he relax — and reveal himself. He was not, he told her, a regular cop. He said he was a special agent with the Federal Bureau of Investigation (FBI), working on a human trafficking and drug trafficking project, and that when he scanned her plate he could see she had “suspicious travel patterns.” In a detail that still reads as faintly absurd, he told her he had doubted her New Orleans story because his system showed no hits there — as if a gap in a surveillance database were evidence that a traveller was lying, rather than evidence that the database is incomplete.

“I was shaken up,” Knowles told the ACLU. “He could tell that maybe I didn’t believe him, so he actually invited me to come into his car, so he could show me what he was seeing — showing me everywhere I’d been.” What she saw, she said, was “basically a heat map of the US, and I could see different circles where I’d been, and when he clicked on it, it was a photo of my license plate.”

When she asked how anyone could be seeing her plates at all, the officer’s answer was blunt: “Well they’re just everywhere.” He said cameras sit on top of highway signs — “they’re often on the back of those, and people don’t even see them.”

Two things about this account need stating plainly. First, it is Knowles’s recollection, as told to the ACLU; no agency has publicly confirmed that the officer worked for the FBI, and the ACLU does not name the software involved or the state where the stop happened. Second, the strangeness of the encounter is precisely the point. Most drivers flagged this way will never be invited into a patrol car to inspect their own heat map. They will simply be stopped, questioned, and sent on their way — or searched — without ever learning that an algorithm, not an officer’s eyes, picked them out.

From looking up a plate to inventing a suspect

To understand why the ACLU treats this as a turning point, you have to understand what license plate readers used to do. A camera photographs a passing plate, logs the plate, location and time, and police later search those records when a particular vehicle is already of interest — a stolen car, a suspect’s car, a car seen near a crime. That is intrusive enough, given the millions of innocent drivers swept into the database along the way. But it at least starts with a human reason to look.

The shift the ACLU describes reverses that order. The database comes first; the reason comes second. AI tools churn through oceans of movement records, looking for patterns their training data labels suspicious — a quick turnaround trip, a route through a “source” state, vehicles repeatedly seen together — and then hand police a name, a plate, or a car to stop. The officer still needs a reason to switch on the lights, and that is where the pretext comes in: a lane drift, a missed signal, a dangling air freshener. In the trade, officers call these “whisper stops” or “intel stops,” because the real reason is whispered from a distant data feed and walled off from the driver, the report, and sometimes the courtroom.

Jay Stanley, a senior policy analyst with the ACLU’s Speech, Privacy, and Technology Project and the author of the 26 October post, says he first noticed the plate-reader company Flock Safety touting this capability in July 2025 — boasting that it could alert police to drivers whose travel patterns were “suspicious” and track associations between vehicles seen together. “More and more people will be subject to police encounters as these AI algorithms inevitably sweep in far more innocent people than actual wrongdoers,” Stanley wrote. “For many people, those encounters will be far more fraught, frightening and dangerous than what Knowles experienced.”

That second sentence matters, and it is worth pausing on. Knowles’s encounter ended with a demonstration and a pamphlet. For a driver who does not look like Knowles, who does not speak as calmly, or who is stopped at night on a quiet highway, the same flag can end with a search, a dog, a drawn weapon, or an arrest. The algorithm does not know any of that. It only knows the pattern.

The ACLU is not alone in documenting the pattern. In November 2025, the Associated Press reported that the US Border Patrol was monitoring millions of drivers nationwide for “suspicious” travel patterns, drawing on plate readers run by the Drug Enforcement Administration, private vendors including Flock Safety, and local agencies, and passing tips to local police who then made pretext stops. One furniture-moving company driver, stopped about 100 miles from the Mexican border after driving from South Carolina, had his truck searched, was arrested on suspicion of money laundering over cash he was carrying, and ultimately faced no criminal charges. The ACLU notes it is possible Knowles was stopped under that programme — possible, not proven.

The same logic shows up in smaller, stranger cases the ACLU cites. In March 2026, The Markup reported that a grandmother in California was interrogated by Customs and Border Protection officers about how much time she spends at local casinos, apparently based on her driving patterns. “It just seems suspicious,” she reported they told her. In April 2025, 404 Media reported a man targeted for a stop because he drove back and forth between his home state of Wisconsin and Michigan — which local police noted was “a known source State for Marijuana as it is legal there.” An entire neighbouring state, treated as a suspicious destination. If that is the standard, then ordinary geography becomes probable cause, and every driver near a state line is a suspect waiting for a camera.

A dollar a minute: the Texas verdict

If Knowles’s story is the warning, the Texas case is the receipt. On 28 September 2026, a federal jury in San Antonio found Bexar County liable for violating the Fourth Amendment rights of Alek Schott, a Houston-area oilfield equipment salesman, after a March 2022 traffic stop on Interstate 35. The jury awarded him $76 — one dollar for every minute he was detained. Seventy-six minutes, $76. The sum was deliberately symbolic, and the symbolism is devastating: that is what an unconstitutional detention was priced at, minute by minute, in a federal courtroom.

Schott had been driving home to Houston from a work trip near Carrizo Springs when his truck was flagged. Bexar County Deputy Joel Babb pulled him over, saying Schott had crossed the road’s fog line. Evidence at trial, including Schott’s dashcam footage, showed he had stayed within his lane, according to the Institute for Justice (IJ), which represented him. Schott was detained for 76 minutes, questioned in the front seat of a patrol vehicle, and subjected to a drug-dog search of his truck. No contraband was found.

“I was returning from a work trip. Just driving back, not realizing there was an entire surveillance network for license plate readers, just marking people to be targeted on their journeys home,” Schott said, in an account carried by the case record. “So, I was just traveling back home, trying to get back to my family and my kids and spent 76 minutes detained by the side of the road for doing nothing wrong.”

The Institute for Justice says it will ask the court for an injunction to stop the county using unsupported traffic stops to facilitate warrantless vehicle searches, and the ACLU notes the organisation’s claim that the same county subjected thousands of other drivers to the same routine. That claim — thousands — is the Institute’s, not a court finding about every one of those stops, and it should be read as such. But the jury’s finding on Schott himself is now a matter of record, and it answers the question the ACLU poses in the starkest possible terms: yes, a surveillance flag plus an invented traffic excuse can add up to a constitutional violation.

One honest complication belongs here, because it illustrates how hard these cases are to pin down. Accounts differ on who tipped off the local police about Schott. The ACLU’s post says a federal fusion centre alerted local police that Schott was suspicious because he had driven from Houston to Carrizo Springs and back the next day. The Associated Press’s earlier reporting on the Border Patrol programme describes Schott’s stop as following a Border Patrol tip. The Institute for Justice’s own trial account describes the tip arriving through a law-enforcement WhatsApp group, from a man the deputy knew only as “Kiki,” whom he believed worked at a federal fusion centre in Laredo. Those accounts may overlap — a fusion centre and a federal programme are not mutually exclusive sources — but they are not identical, and a reader deserves to know that the chain of custody for the tip is disputed rather than settled. What is not disputed is the outcome: the stop, the 76 minutes, the empty search, and the jury’s verdict.

That dispute points to a deeper problem for anyone who wants to challenge a stop like this. Defendants face steep barriers to getting the underlying data and methodology: which cameras saw the car, which algorithm scored the trip, what threshold counts as “suspicious,” and whether a human ever checked the flag before acting on it. If the prosecution will not disclose the system, and the vendor calls it proprietary, the driver is left arguing against a black box that has already decided she looks guilty. A legal system that cannot examine its own evidence is not a legal system; it is a rumour with handcuffs.

The machine behind the flag: Flock’s OS Investigate

The most concrete look inside that black box came not from a courtroom but from Flock Safety’s own website. In August 2026, WIRED reconstructed a new Flock AI platform called OS Investigate — previously developed under the codename Nightshift — from more than 450 source files served by Flock’s login portal. The picture it painted was very different from the company’s long-standing assurance that its technology “cannot recognize, identify, or track individuals.”

Photo: An automated license plate reader system on the south side of eastbound Interstate 10 near Vail, Arizona, in March 2025. File photo — it shows the kind of roadside system at issue, not a system tied to any individual case in this article. Credit: Dugan Meyer, CC BY 4.0, via Wikimedia Commons.

OS Investigate, WIRED reported, ships with 69 prewritten prompts officers can select, edit and submit to the AI. Nineteen focus on patterns, and 14 require no plate, name or physical description at all — just a place, a time window and a behaviour. One prompt asks for “witnesses based on vehicles most seen in [neighborhood] during [last 14 days].” Another seeks vehicles that visited three or more shops in three days. Another asks for everyone arrested more than twice in two years for non-drug offences, then builds dossiers on the top three. Beyond camera scans, the code showed the system could pull in police case files, arrest records, 911 dispatch logs, ballistics results and commercial identity databases.

Here the two sides diverge, and both claims belong in the same paragraph. The ACLU’s October post describes WIRED’s system as already in use by some police. Flock, in response to WIRED, said the tool was still in development and in testing with a small group of law enforcement partners. Those statements are not quite contradictory — a small testing group is still “some police” — but they point in opposite directions about scale, and scale is everything here. A tool in testing can be contained. A tool in use is already shaping stops.

Chad Marlow, an attorney at the ACLU, told WIRED that the prompt design itself was the danger: “And I think the fact that they don’t seem to have any limitations on the prompts enables the police to almost lead the AI in what it does. The difference between a police officer saying, ‘Do you see any criminal patterns’ or ‘Find me criminal patterns’ is going to produce two different things.” He is right, and the distinction is not academic. Ask a system whether it sees a crime and you invite caution. Tell it to find one and you have ordered it to hallucinate a suspect.

We checked the ACLU’s original post against Gizmodo’s follow-up coverage, the Institute for Justice’s trial account and WIRED’s reporting as summarised by multiple outlets, and the gaps are consistent across all of them: nobody outside the vendors and the agencies can currently audit what counts as a suspicious pattern, how often flags are wrong, or how many innocent drivers are stopped for each genuine find. That opacity is not a side effect. It is the product.

Can a road trip be reasonable suspicion?

Here is where this article stops reporting and starts arguing, because the law here is not actually that complicated, and pretending otherwise helps no one. The Fourth Amendment requires reasonable suspicion, based on specific and articulable facts, before police can stop you — and a pattern is not a fact about you. It is a statistical resemblance to other people, computed from incomplete camera coverage, by a proprietary model, on data you never consented to create. A quick trip to Carrizo Springs and back is a fact about a route. It is not a fact about a crime. A road trip through three states is a fact about a holiday. It is not a fact about trafficking. Treating either as suspicion collapses the difference between “this looks like something criminals do” and “this person is doing something criminal” — and that difference is the entire point of the Fourth Amendment.

The counterargument deserves a fair hearing: police have always used patterns. A detective who notices the same car circling a block is using pattern recognition, and nobody thinks that is unconstitutional. But there is a colossal difference between a detective’s observation, which can be cross-examined in court, and an algorithmic flag, which often cannot. The detective can be asked: what did you see, when, and why did it matter? The algorithm answers, in effect: trust me. And when the flag is wrong — when the New Orleans visit never registered, when the lane drift never happened, when the stolen-car alert was one of 161 false alarms — the driver pays for the error in minutes, dignity and sometimes freedom, while the system learns nothing, because nobody was watching it learn.

That is why the Schott verdict matters beyond its $76. A jury did not merely find that one deputy behaved badly. It found that a county policy or custom — a routine, repeated way of doing business — caused the violation. Routines do not reform themselves. They have to be stopped, by courts, by legislatures, or by the public refusing to accept that a heat map is a warrant.

There is a shoddy piece of logic at the centre of all this that deserves naming: the idea that because a system watches everyone, it therefore suspects no one in particular, and so no one’s rights are violated until the stop. That is backwards. Mass surveillance followed by algorithmic selection is not the absence of suspicion. It is suspicion applied to everyone, all the time, with the actual stop deferred until the machine picks a winner. Living under that arrangement — knowing that every trip to a clinic, a protest, a casino, a gun shop or a lover’s house is being scored — is itself a chilling of free movement. You do not need to be stopped to change how you drive. You only need to know you might be.

The backlash is already bipartisan

If there is good news in this story, it is that the backlash has outgrown the usual partisan script. An August 2026 YouGov survey shared with The Washington Post found 46% of respondents opposed police license plate cameras in their communities, against 38% who supported them — a reversal from a year earlier, when a majority were in favour. That is not a fringe position anymore. It is the plurality position.

The policy response has followed, unevenly but unmistakably. In Los Angeles, the Police Department let its three-year Flock contract expire in July 2026 after its own inspector general found 161 stolen-vehicle alerts in a two-month sample were wrong — roughly one in three alerts in the period reviewed — each capable of triggering a high-risk stop of an innocent driver. In Florida, Governor Ron DeSantis’s administration ordered automated license plate readers removed from state highway rights-of-way. In Texas, Governor Greg Abbott paused state funding for Flock cameras in late August 2026, after reporting that a state authority had directed at least $30 million toward more than 3,200 Flock cameras; the pause did not remove existing cameras, and how long it lasts remains unclear. In Pennsylvania, Governor Josh Shapiro said he would welcome bipartisan legislation to ban the cameras statewide. In Missouri, Senator Josh Hawley opened an investigation into Flock Safety in August 2026, demanding documents from chief executive Garrett Langley. Taken together, these moves are a sign that, for once, the left’s fear of a surveillance state and the right’s fear of an unaccountable federal database are pointing at the same cameras.

None of that, on its own, fixes the problem Knowles ran into. A funding pause is not a ban. A lapsed contract is not a deletion of the data already collected. A bill is not a law. The question now is whether the rules that emerge will actually require what matters: warrants for historical searches, short retention limits, public audits of flag accuracy, and a right for a stopped driver to see — and challenge — the algorithmic reason she was stopped. Without those, the next Knowles will still be handed a pamphlet test she never asked to take.

For now, the practical advice is bleakly simple, and it is the same advice the ACLU’s post implies without quite stating: assume you are being recorded when you drive, because you are. Assume a pattern could be built from your trips, because it can. And if you are stopped for a trivial reason that feels invented, remember Schott’s 76 minutes and his $76 — and remember that a jury, given the dashcam and the timeline, believed him over the excuse. The road trip is not the crime. The system that treats it as one is the story. And until that system can explain itself in a courtroom, in plain language, to the person it flagged, no driver should accept a heat map as an answer to the question: why me?

Daniel Rao covers technology, AI policy and civil liberties for TechPulse USA. He writes about the systems that watch us back.

Sources: American Civil Liberties Union, “Police Are Increasingly Pulling People Over Because AI Deems Their Driving Patterns ‘Suspicious’” (26 October 2026, Jay Stanley); Gizmodo, “Police Are Using AI to Pull Over Drivers for ‘Suspicious’ Behavior” (27 October 2026); WIRED’s August 2026 reporting on Flock Safety’s OS Investigate, as reported by WIRED and summarised by TechTimes, Decrypt and Cato; Associated Press investigation into the Border Patrol’s driver-monitoring programme (November 2025); Institute for Justice / Police1 trial accounts of Schott v. Bexar County (verdict 28 September 2026); The Markup (March 2026) and 404 Media (April 2025) cases cited by the ACLU; YouGov survey (August 2026) via Gizmodo and TechCrunch; Texas Tribune / Governing on Governor Abbott’s August 2026 funding pause; Los Angeles Police Department inspector general audit via TechTimes and Smart Cities Dive; Pennsylvania Governor Shapiro via The Center Square / GoErie; Senator Hawley’s August 2026 investigation letter via Smart Cities Dive.

Related on TechPulse USA: Hackers Tore Down a Flock Camera, Cracked Its Encryption Key, and Exposed How It Really Tracks You; Savannah Officer Gave Her ICE Husband Her Flock Login — He Ran 983 Searches; Norway Wants Meta’s Camera Glasses Banned From Public Spaces.