Informopedia Explains: Grace Knowles was between jobs, in her twenties, driving across America. Her route took her through Oregon, Texas and Louisiana, and on a morning in September 2025 she left New Orleans heading for Florida. Somewhere on the road, an officer pulled her over and told her she had spent too long driving in the left lane.
Knowles disputed that. The questions that followed were not about her driving. They were about her itinerary: where she had come from, where she was going, whether she could prove she had been in New Orleans. She showed the officer a pamphlet from a whiskey museum she had visited. He relaxed — and then explained, according to her account to the American Civil Liberties Union (ACLU), that her licence plate had been flagged for suspicious travel patterns. He showed her his car’s computer: a heat map of the United States, circles marking the places she had been, and, when he clicked one, a photograph of her licence plate.
Asked how those images had been collected, the officer told her the cameras were everywhere — including on top of highway signs.
Knowles’s story, published by the ACLU on October 26 in a post by senior policy analyst Jay Stanley, is the human face of a technical shift that matters to every driver: licence-plate readers no longer just help police find a stolen car. Combined with AI, they can now generate suspicion about people the police had no reason to suspect at all.
Step one: the camera reads your plate
An automated licence plate reader (ALPR) is a camera — mounted on a police car, a pole, a highway sign or a bridge — that photographs passing vehicles and extracts the plate number, the location and the time. One camera gives one data point. Thousands of cameras, operated by different agencies and private companies, give a trail: where a vehicle has been, how often it visits a place, which other vehicles it travels near.
The companies that sell these systems, notably Flock Safety and Axon, market them as crime-fighting tools, and in that role they work: a stolen car, an Amber Alert vehicle, a car linked to a known suspect can be found quickly. The data is also shared widely between agencies, and often retained for months or years, which means a record of ordinary travel accumulates whether or not its owner is ever suspected of anything.
That accumulation is the raw material for step two.
Step two: the software looks for patterns
The shift Stanley describes is from searching to flagging. In the old model, police started with a suspect vehicle and searched plate records for it. In the new model, AI systems search enormous movement records for patterns they have been trained to treat as suspicious — recurring routes, unusual destinations, apparent links between vehicles — and generate alerts about vehicles nobody was investigating.
Stanley says he first saw Flock pitching AI that alerts police to “suspicious” travel patterns in July 2025. The ACLU’s warning is blunt: “More and more people will be subject to police encounters as these AI algorithms inevitably sweep in far more innocent people than actual wrongdoers.” A cheap road trip between jobs, a driving route shaped by budget motels and fuel prices, a visit to a clinic, a place of worship or a protest — all are patterns to an algorithm, and the algorithm does not know why you drove.
Step three: the pretext stop
An algorithmic flag is not, by itself, a legal reason to stop a car. So the flag needs a cover story — and that is where the “pretext stop” comes in. An officer cites a minor traffic offence, real or invented, as the official reason for the stop, while the real reason is the surveillance data. In Knowles’s case, the stated reason was lingering in the left lane. In other cases documented by the ACLU, officers have concealed the actual reason entirely.
Within law enforcement, these encounters are sometimes called “whisper stops” or “intel stops”: a tip arrives quietly — from a fusion centre, from a federal agency, from an analyst — and a local officer makes the stop for speeding, a missed turn signal, or a dangling air freshener. The driver never learns that software chose them.
That opacity is also a legal problem. A driver who wants to challenge a stop needs to see the data and the method behind the flag — and defendants face steep barriers to obtaining either. The ACLU’s broader argument is constitutional: accessing a person’s location history without a warrant is a search, and it should require one, except in emergencies.
What the courts and states are already doing
The argument is not theoretical. On September 28, a federal jury in San Antonio found Bexar County liable over a traffic-stop programme that used plate-reader data, in a case brought over a March 2022 stop of Alek Schott on Interstate 35 near San Antonio. Schott was held for 76 minutes. The jury awarded him $76 — a dollar a minute — and his lawyers at the Institute for Justice are now seeking an injunction against the county’s stops. Accounts differ on who raised the original alarm: the ACLU says a federal fusion centre tipped off local police, while other reporting attributes the tip to Border Patrol agents. That disagreement itself shows how hard these stops are to scrutinise.
In Wisconsin, the ACLU has gone further, petitioning the state Supreme Court directly on behalf of the Wisconsin Council of Churches. Its argument: residents cannot drive to a church, a clinic, a protest or a recovery meeting without being photographed and logged, and accessing that history without a warrant violates the state constitution’s ban on unreasonable searches. Churches are the chosen plaintiffs for a reason — they run addiction recovery programmes, domestic-violence support and political gatherings, all activities whose participants have good reason to expect privacy.
And the politics are shifting under the technology. Los Angeles police have suspended or ended contracts with Flock Safety; Florida has revoked local law enforcement’s ability to place plate readers on state highways; Texas has paused state funding for Flock cameras. The backlash crosses party lines, because the objection is not to catching criminals — it is to a system that watches everyone first and asks questions later.
What we do not know — and why it matters
Honesty requires noting what the ACLU’s account does not establish. It does not name the officer who stopped Knowles, the software that flagged her, or the state where the stop happened, and it does not establish whether an AI system generated her flag at all. It does not count how many AI-flagged stops have occurred, or how often they find wrongdoing versus inconvenience the innocent. The ACLU post also does not include responses from Flock or the FBI. Those gaps are not footnotes: without them, nobody can say how widespread the practice is, or how a driver could challenge a stop built on an algorithmic judgement.
But the direction of travel is clear, and it connects to a wider surveillance story — from an Israeli firm turned online ads into a spy operation to hacking tools seized from China’s Flax Typhoon operation. The tools differ; the pattern is the same. Data collected for one purpose acquires a second life, and the people in the data are the last to know.
The ACLU’s answer is the most sweeping available: end licence-plate surveillance and location tracking altogether. That is unlikely to happen. What is more likely — and what the San Antonio verdict and the Wisconsin petition point toward — is a middle path: warrants before police can browse location histories, limits on retention, and disclosure when a stop starts with software rather than observation. Until then, Knowles’s heat map is the best image of the system we have: a map of a country, covered in circles, each one a place somebody went, kept by cameras they never noticed, waiting for an algorithm to decide it looks suspicious.
