For decades, a car's number plate was little more than an identifier.
Police could photograph it. A camera could record it. An officer could check it against a database.
But what if a computer could reconstruct where your car has been, who it may have travelled alongside, and the places it repeatedly visits — without you ever being suspected of a crime?
That possibility is becoming increasingly real.
A new generation of artificial intelligence is transforming automated number-plate recognition from a system that finds specific vehicles into something potentially much broader: a searchable map of vehicle movements.
And one of the most controversial examples is being developed by US surveillance technology company Flock Safety.
From reading plates to understanding patterns
Traditional automatic licence plate recognition, or ALPR, is relatively straightforward. A camera sees a vehicle. The system reads its registration number. The time and location are recorded.
Police can then search the database when investigating a particular incident.
Flock says its cameras are designed to capture vehicles and visible characteristics rather than identify people, and argues that its ALPR systems are not capable of tracking individuals over time.
But artificial intelligence changes what can be done with the accumulated data.
According to an investigation published by WIRED in August 2026, Flock has developed an AI system called OS Investigate, previously known as Nightshift, that goes considerably beyond simply searching for a known registration number.
Instead of starting with:
«“Where has this vehicle been?”»
an investigator could potentially start with a location, time period or behavioural pattern and ask the system to find vehicles matching those characteristics.
The database becomes the investigator
WIRED examined code associated with the system and reported that OS Investigate contains dozens of pre-written prompts for police users.
Some searches can reportedly look for vehicles repeatedly visiting particular locations or travelling between particular areas. Other functions can identify vehicles that frequently appear near another vehicle.
In one example described by WIRED, the system could analyse vehicles appearing near a target vehicle within a defined time window and identify potential “associates” based on repeated co-occurrence.
That creates a fundamentally different type of surveillance. The system isn't necessarily looking for your car. It can potentially look for patterns involving cars.
And you could become part of the resulting pattern without ever being the original target.
Imagine the system knows your routine
Consider an entirely hypothetical example.
Your car leaves home every morning at 7:30. It appears on the same road. It is repeatedly detected near your workplace. On some evenings it travels to a particular shopping area. At weekends it frequently appears at another location.
Individually, none of those observations is especially revealing. But collect them over months and an algorithm could potentially construct a remarkably detailed picture of a vehicle's routine.
The technology doesn't necessarily need to know who is sitting behind the wheel. The movement pattern itself becomes information.
And if the vehicle can subsequently be connected to an individual through other records, the distinction between tracking a car and tracking a person becomes considerably less meaningful.
The most controversial step: searching without a suspect
This is where the technology becomes particularly significant. Traditional policing generally begins with something concrete.
A crime occurs. There is a location. There may be a witness. There may be a vehicle description. Investigators then search available evidence.
AI-driven systems potentially reverse that process.
According to WIRED, some of the prompts examined in OS Investigate require no specific registration number, name or identified suspect. Instead, an officer can provide a location, timeframe and behavioural pattern and ask the system to identify vehicles fitting those criteria.
That creates the possibility of pattern-based investigation. The system doesn't necessarily begin with: “Find this suspect.”
It can begin with: “Find people who behave like this.”
That is a profound shift.
From surveillance to predictive policing?
This raises an uncomfortable question. At what point does searching historical evidence become predicting suspicious behaviour?
Suppose an AI identifies a vehicle because it repeatedly visits several locations late at night. That behaviour might be completely innocent. Perhaps the driver works nights. Perhaps they are a delivery driver. Perhaps they care for a relative.
The algorithm doesn't inherently understand the reason. It just identifies a pattern.
This creates the possibility of false suspicion being generated by the system itself.
Privacy advocates have therefore raised concerns about surveillance systems that allow searches to begin with behavioural patterns rather than specific suspects. WIRED reported that experts reviewing the system described this as moving toward a much broader form of investigative surveillance.
Your car could have “associates”
One of the more striking concepts is the idea of vehicle association. If two cars repeatedly appear near one another, an algorithm can potentially determine that they are connected.
But connection does not necessarily mean relationship. It simply sees a statistical pattern.
That pattern can then become an investigative lead.
The information can become personal
Vehicle surveillance becomes even more powerful when combined with other databases.
According to WIRED, OS Investigate can potentially work across multiple categories of information available to participating law-enforcement agencies, including police records, dispatch information and commercial identity databases.
That means a vehicle record can potentially become much more than a vehicle record.
A registration number can be associated with an owner. An owner can have an address. An address can be associated with other people.
Other records can potentially reveal telephone numbers, relatives or previous interactions with law enforcement.
Suddenly, a photograph of a passing car has become the starting point for a much larger investigation.
Scale is the real issue
A single camera isn't particularly revolutionary. The significance comes from scale.
Flock says its network now spans thousands of communities, while WIRED reported that the company's cameras collectively log billions of plate scans each month.
At that scale, individual observations become part of a vast historical dataset. The system doesn't necessarily need to follow your car continuously. It can potentially reconstruct parts of its journey from repeated observations made by different cameras.
The result is something resembling a fragmented travel history.
Seven days — or much longer?
Data retention is therefore critical.
Flock announced in August 2026 that it was recommending a seven-day default retention period for ALPR data, alongside measures intended to detect misuse and strengthen accountability.
That is significant because retention determines how much historical information can potentially be reconstructed.
A seven-day dataset is very different from a six-month dataset. A six-month dataset is very different from several years.
The longer information exists, the more behavioural patterns can potentially be extracted from it.
The company says there are safeguards
Flock disputes the idea that its technology should simply be characterised as mass surveillance.
The company says its ALPR cameras capture point-in-time observations of vehicles in public view and argues that they are designed for investigative purposes rather than monitoring individuals.
Flock has also announced additional safeguards, including mandatory misuse detection, case-code requirements, proactive lockouts and stronger authentication measures.
The company also describes newer AI tools as ways for investigators to find relevant evidence more efficiently, including searches based on natural-language descriptions rather than conventional database fields.
There is a legitimate argument behind this.
If a serious crime has been committed, investigators may have only fragments of information. A witness might remember the colour of a vehicle but not its registration. A camera might contain useful footage that would otherwise take hours to locate.
AI could dramatically reduce that search time.
But efficiency isn't the same as restraint
The central privacy question isn't necessarily whether the technology can solve crimes.
It is whether the ability to search enormous quantities of movement data should be constrained by strong rules.
A powerful search system can be extremely useful when investigators have a legitimate target. It becomes more controversial when the system can generate targets from patterns.
The distinction is subtle but crucial.
Finding evidence about a suspect is one thing. Finding suspects from everyone's historical movements is another.
Britain is watching the same technological shift
The debate isn't limited to America. Britain already has extensive CCTV infrastructure, automatic number-plate recognition and increasingly sophisticated data-analysis capabilities.
The UK therefore has many of the ingredients required for increasingly automated vehicle intelligence.
The crucial questions are likely to revolve around how different datasets can be connected, how long information is retained, who can access it and what constitutes a legitimate search.
As AI becomes better at finding relationships within large datasets, the practical limits on surveillance may increasingly depend less on what the technology can do and more on what organisations are allowed to ask it to do.
What happens when AI gets it wrong?
This may ultimately be the biggest danger. Humans make mistakes. AI systems make mistakes too.
But an AI-generated pattern can look remarkably authoritative. A list of vehicles. A map. A timeline. A probability score. A network showing connections.
The presentation itself can create an impression of certainty. Yet a statistical association is not proof of wrongdoing.
A car appearing near another car doesn't prove the occupants know one another. A vehicle visiting several locations doesn't prove criminal activity. A person travelling frequently at night isn't inherently suspicious.
The more powerful the system becomes, the more important it becomes to distinguish correlation from evidence.
The end of anonymous driving?
For much of the twentieth century, driving provided a degree of anonymity.
You could leave your house, drive across a city and arrive somewhere without creating a permanent digital diary of every road you travelled.
That world is disappearing.
Sat-nav systems, smartphones, connected vehicles, toll systems, parking cameras and ALPR networks can all generate information about journeys.
AI can then connect those fragments. The result is potentially something much more comprehensive than any individual surveillance system.
Not one camera watching you. Thousands of independent observations that can be assembled into a picture of your movements.
The car becomes a digital footprint
The vehicle sitting outside your house may appear to be nothing more than a machine. But in a connected surveillance environment, it can become an identifier.
Its registration number can link observations together. Its movements can reveal routines. Its repeated encounters with other vehicles can reveal associations.
Its presence at particular places can create a timeline. And artificial intelligence can make sense of all of it far faster than a human investigator could.
That is why this technology matters.
The issue isn't simply that cameras can read number plates. They have been doing that for years.
The more profound development is that AI can potentially turn millions of individual observations into a searchable model of movement and behaviour.
The road ahead
Vehicle surveillance is unlikely to disappear. There are compelling reasons for police and emergency services to use it.
Finding stolen cars, locating missing people, investigating hit-and-runs and reconstructing the movements surrounding serious crimes can save time and, potentially, lives.
But the same technology can also create unprecedented visibility into ordinary people's movements.
The challenge will be deciding where the line is drawn.
How much historical movement should be searchable?
How specific must a police investigation be before an AI system can be used?
How should innocent people be protected from being swept into searches?
How long should movement records exist?
Who audits the searches?
And perhaps most importantly:
Should an algorithm be allowed to decide which patterns look suspicious in the first place?
The technology is moving rapidly.
Flock's latest AI systems illustrate where vehicle surveillance could be heading: away from simply reading licence plates and towards systems capable of searching relationships, behaviours, locations and patterns across enormous datasets.
Your car may not know where you've been. The cameras don't necessarily know who you are.
But when artificial intelligence connects enough observations together, someone — or something — may be able to reconstruct a surprisingly detailed map of your journey.
And the most unsettling part is that you may never know that your car was part of the search.
Comments
Post a Comment