A number plate can be used to find a stolen car. Linked to a series of places and dates, it can also tell the story of a journey. The shift between these two uses lies at the heart of a warning published on 26 October by the American Civil Liberties Union (ACLU): the US civil liberties organisation describes road checks linked to the analysis of movements deemed suspicious.
The issue goes beyond the automatic reading of a number. It concerns the decision made on the basis of that number: whether to look for a vehicle already identified in an investigation, or to regard a route as deserving police attention in its own right. To understand the difference, we need to look separately at the testimony, the capabilities of the tools and the safeguards governing their use.
An account of a road check, with precise limitations
The ACLU reports the testimony of Grace Knowles, a young American woman stopped during a trip in September 2025. According to her account, the officer allegedly told her that her movements were suspicious and showed her a map accompanied by photographs of her number plate.
The testimony was collected and published by the organisation. On its own, it does not constitute an audit of the software used. The article does not make it possible to attribute the stop with certainty to a specific AI model, or to establish that Flock Safety was the supplier involved in this incident. Conflating the account with a technical demonstration would imply a degree of certainty that the available evidence does not provide.
The ACLU takes broader issue with the use of number-plate databases to generate suspicions based on people’s movements. This is a civil liberties position advanced by an organisation active in this area. It deserves to be considered alongside the functions actually advertised and the conditions under which the systems are used.
What the manufacturer says it offers
On its page about automated number-plate reading, Flock Safety describes cameras capable of capturing a number plate and vehicle characteristics such as its make, colour or body type. The company presents several uses: finding a stolen vehicle, searching for a missing person or matching observations collected across several jurisdictions.
The platform also advertises alerts for vehicles on watchlists and, with FreeForm, searches made using natural language. Flock says its offering does not include facial recognition and that every search is recorded. These details describe what the supplier offers; on their own, they do not measure the actual effectiveness of a road check or the frequency of errors.
One technical distinction remains essential. Reading a number plate correctly and interpreting a journey correctly are two different tasks. Even if the number plate has been recognised without error, the meaning attributed to the movements still needs to be justified. A tool can provide an accurate observation and nevertheless lead to a weak hypothesis.
Three decisions behind an alert
To assess a system of this kind, its use can be broken down into three stages. First, collection: which vehicle was observed, where and when? Next, analysis: what link is established between the observations? Finally, action: who decides that they justify a check or intervention?
This breakdown avoids talking about an AI that “stops” a driver as though the software carried out the entire process on its own. The technology selects, matches or presents information; people define how it is used and make the decisions that follow. Human responsibility remains, even when an interface makes an alert look precise.
Consider a hypothetical example, unrelated to any police case: two vehicles regularly travel along the same road at around the same time. This coincidence could reflect a commuting route shared by thousands of people. More evidence would be needed to see it as a meaningful association. A collection of points on a map does not automatically explain them.
The safeguards promised must be open to scrutiny
In its presentation on trust and privacy, Flock says that access is granted to users authorised by its customers, that searches are linked to a user and that they are recorded. The company also says that sharing between organisations depends on a decision to opt in. It presents its system as a targeted investigative tool, rather than as continuous general surveillance.
These commitments raise questions that can be checked: what grounds are accepted for a search? Who reviews the access logs? How is unjustified use detected and dealt with? A record makes it possible to trace an action; there still needs to be a procedure for reviewing it.
The debate therefore cannot be reduced to a choice between a commercial promise and general concern. A serious assessment should examine permitted uses, recorded errors, human checks and opportunities to challenge decisions. The pages consulted do not provide an independent measure of the rate of unjustified interventions. The absence of such a measure rules out both claiming perfect reliability and arbitrarily quantifying abuses.
The decisive point remains the transition from information to decision. A number plate and a route can help an investigation. They do not remove the need to explain why someone becomes a suspect.
Sources
- ACLU, Jay Stanley: testimony and analysis of road checks, 26 October 2026.
- Flock Safety: advertised features of its number-plate-reading platform.
- Flock Safety: commitments on access, searches and data sharing.
Illustrative photo: a Flock camera at the intersection of East San Salvador Street and South 4th Street in San José, in October 2026. Eirene12 / Wikimedia Commons, CC BY 4.0.
