Flock’s New AI Police Tool Can Identify Drivers, Reconstruct Routes, and Profile People From Movement Patterns

Flock Safety has developed an artificial intelligence system for police that can identify drivers, infer relationships between vehicles, and turn routine camera data into suspect lists, witness leads, and personal dossiers, according to code reviewed by WIRED. The product, now being tested under the name OS Investigate, marks a major expansion of the surveillance company’s capabilities and raises fresh constitutional and civil-liberties concerns.

The tool matters because it appears to shift license-plate readers from a narrow “find this car” function into a broader system for mapping where people go, who they travel with, and who might fit a search based only on location, timing, or physical description. That could make police investigations faster, but it also deepens worries about dragnet surveillance and warrantless tracking.

For years, Flock Safety has marketed its cameras as a way to check plates against hotlists and ignore vehicles that are not of interest. The new system, however, is described in code as something much more ambitious: a police-facing AI layer that can query camera networks, records databases, and commercial identity services to produce names, addresses, relatives, and behavior-based leads.

What WIRED found in Flock’s code

WIRED says it examined software files served from Flock’s own login pages and found evidence of an AI product with 69 preset prompts for police users. Those prompts, along with surrounding code, suggest officers can ask the system to search for witnesses, identify “associates” of a driver, run background-style “workups,” and locate people who match a physical description inside a mapped area.

The reporting indicates the software was originally called Nightshift before being renamed OS Investigate. Flock told WIRED the product is still in development and being tested with a small group of law-enforcement partners, and that its features may change before any broader release.

Even in its current form, though, the code appears to describe a system with access to a wide range of police and commercial data sources. That includes plate scans, camera metadata, arrest records, case files, 911 dispatch logs, ballistics data, and identity records containing information such as Social Security numbers, birth dates, phone numbers, email addresses, and family connections.

How does OS Investigate differ from a standard license-plate reader?

It differs by treating a recorded vehicle not just as a data point to be checked, but as a clue to be analyzed for movement habits, associates, and likely identities. In Flock’s traditional model, a camera captures a plate, compares it with a list of wanted vehicles, and ignores everything else.

OS Investigate appears to invert that model. Instead of starting with a specific plate or named suspect, officers can begin with a place, a time range, or a pattern of behavior. The system is then designed to identify the people who fit those conditions, even if no crime has yet been linked to them.

That shift is at the center of the controversy. Privacy advocates say a tool built this way does not merely help police search existing records; it can create new suspicion by drawing in people based on routine travel patterns, repeated visits to locations, or proximity to others.

What kinds of searches are preloaded?

The code reviewed by WIRED includes a menu of canned prompts that officers can select and edit before submitting. Some are direct, such as searches for witnesses whose vehicles appear most often in a neighborhood over a set period. Others are more expansive, asking for people with repeated arrest histories, then mapping where they live and pulling related service calls.

Several prompts focus on pattern detection rather than record lookup. Examples described in the reporting include searches for vehicles that visit multiple retail sites in a short span, pass through several cities in a week, or appear at a series of gas stations during overnight hours. In many cases, the system does not need a plate number, a name, or an incident report to begin.

One of the most consequential details is that the software can strip out certain categories of vehicles, such as buses, semi-trucks, work vans, and trailers. That means a delivery worker making a series of stops could be filtered away, while a personal vehicle following the same path might be retained and flagged.

Why are privacy experts alarmed?

Privacy experts say the concern is not simply that police can search more data, but that the software lowers the threshold for starting a search in the first place. Instead of requesting information tied to a known suspect, an officer can cast a wider net and let the system decide which drivers, patterns, or associates are worth attention.

That approach, critics argue, pushes police toward suspicion by algorithmic suggestion. It also blurs the difference between identifying a vehicle associated with a crime and building a profile of a person whose travel habits happen to fit a search template.

Jay Stanley of the ACLU’s Speech, Privacy, and Technology Project said the public often thinks of license-plate readers as a simple wanted-vehicle check. In his view, Flock’s growing use of AI has narrowed the distance between what the technology can do and how it is being used in practice.

Stanley argued that the emerging system resembles a far more aggressive surveillance model than most people would expect from a plate reader, warning that the gap between technical capability and real-world use is shrinking fast.

Chad Marlow, who leads the ACLU’s “Get The Flock Out” campaign, said prompt design matters because it shapes how officers ask the machine to investigate. He warned that if the system is optimized to search for “criminal patterns,” it can reflect the AI’s own logic rather than a legal standard of suspicion.

Other experts see constitutional implications as well. The reporting notes that the tool appears capable of combining camera histories with police and commercial records in ways that can recreate a person’s movements without a warrant. That raises questions about whether the system effectively creates a searchable, privately owned geofence around entire communities.

What the prompts reveal about police surveillance

The prompts are important because they show not only what Flock’s customers might be able to ask, but what Flock itself expects them to ask. They also reveal a design philosophy: automate the most common and most expansive investigative steps so an officer can type a few words and get a structured result.

Some prompts are focused on finding witnesses, but others are framed around broad categories of arrest history, repeated visits to commercial sites, or routes between cities. In several cases, the officer supplies only a location, a date range, and a behavioral pattern.

That matters because the system’s power no longer depends only on a direct hit against a hotlist. It can also surface people who are not wanted, not named, and not necessarily under investigation, but who are deemed statistically relevant by the search criteria.

Examples of the search logic

• Find vehicles most often seen in a neighborhood over the last two weeks during specific hours.
• Identify drivers who visited three or more stores in one city over a three-day period.
• List vehicles that entered multiple banks or gas stations within a short timeframe.
• Search for a person matching a physical description near a map-drawn area.
• Generate a “workup” on the top-ranked people returned by the system.

Flock’s terminology for a “workup” is especially notable. The company describes it as a one-step background check that starts with a name and date of birth and returns available departmental records plus commercial data, including vehicles, prior suspect listings, relatives, phone numbers, and online accounts.

That function turns a simple identification task into something closer to an intelligence dossier. In practice, that means an initial search can cascade into a more detailed portrait of someone’s life, family, and associations.

How does Flock say the product should be understood?

Flock says OS Investigate is distinct from its license-plate-reader product and is meant to help investigators work across information their agencies already can access. The company also says the tool is still being tested and that its workflows and capabilities could change before a wider launch.

In a statement to WIRED, company spokesperson Paris Lewbel said the system should be viewed as a separate product and not as the same thing as its core camera network. He added that the features seen in the code may not be identical to what Flock ultimately sells.

That framing is consistent with how the company has long described itself publicly. Flock’s trust and safety materials say its technology is intended for specific investigations, not general monitoring. The problem, critics say, is that the code WIRED reviewed appears difficult to reconcile with those claims.

Flock’s public position, according to the reporting, is that its technology is meant to support targeted cases rather than broad surveillance, but privacy advocates say the internal prompts and data links suggest a much wider reach.

Why this is happening now

The timing is significant because Flock is facing a growing backlash over the ways police have already used its cameras. The company has become a politically charged symbol of local surveillance, drawing criticism from both civil-liberties groups and some lawmakers while continuing to expand into more departments and jurisdictions.

According to the reporting, Flock systems are deployed in more than 6,000 communities and process roughly 20 billion plate scans a month. The company has also raised large sums from major venture firms and reached a valuation of $7.5 billion in its latest funding round.

At the same time, there is mounting evidence that some officers have misused plate-reader tools for personal purposes or overbroad searches. That pattern has made new features especially sensitive, because every added capability increases the stakes of how the system is governed.

Documented controversies around Flock

• An officer in Texas allegedly searched more than 83,000 cameras while looking for a woman who had self-managed an abortion.
• A Washington Post review found dozens of officers charged with or accused of abusing plate-reader systems, including searches tied to personal relationships.
• Illinois regulators found Flock had violated state law in a pilot program that gave Customs and Border Protection access to cameras in the state.
• Separate reporting has shown that many agencies enter vague justifications for their searches, making oversight difficult.

Flock has also said it plans to add more safeguards, including monitoring for abnormal searches, automatically locking out flagged users, and requiring officers to provide a reason for searches. The company says those measures will roll out by the end of the year.

Still, critics note that much of the enforcement burden remains with local agencies. Departments decide who gets credentials, whether camera feeds are shared across jurisdictions, and whether optional anti-abuse features are enabled.

What does the code say about oversight?

The code suggests there are some procedural checks, but they may be more form than substance. According to the reporting, the system requires users to enter a reason for a search by default, yet the form appears to impose no meaningful limit on what that reason can say or how detailed it must be.

If a department requires a case number, the interface appears to check only that it contains at least three characters. WIRED could not determine whether Flock applies more stringent review on the server side after a request is submitted.

That uncertainty is important. A user interface may look restrictive while the underlying system remains broad, or a company may enforce safeguards only after data is already queried. Without transparency into server-side behavior, public oversight remains incomplete.

Independent researcher Buchodi, who examined the code after WIRED shared its findings, reportedly reproduced key parts of the analysis. But even that deeper review could not answer the central question of what happens once a search reaches Flock’s systems.

Who is using this and why might police want it?

Police investigators may want the tool because it promises to compress work that traditionally takes hours or days. According to Flock’s own demonstrations, an officer could search citywide cameras for vehicles linked to an armored-truck tailing incident, then quickly retrieve owners and arrest histories.

For agencies under pressure to solve thefts, find missing people, or identify suspects quickly, that sort of speed is attractive. A California detective quoted in the reporting said the concept would be useful in his department and dismissed objections as overblown, even while acknowledging that some prompts made him uneasy.

But supporters of the technology and critics are talking past each other on a basic question: whether police should be allowed to start with a broad behavioral pattern and let software infer likely suspects. That is where the risk of overreach becomes hardest to defend.

There is also the issue of “association” analysis. The code appears to rank cars by how often they appear near a target vehicle at cameras within a two-minute window, then identify up to 20 possible associates if they appear together enough times. That is a powerful inference engine, not just a records lookup.

How much does the system know about movement?

It knows enough to recreate patterns of life across cities, according to the code reviewed in the report. Searches can look for vehicles that leave one city and return within a week, make repeated round trips within 14 days, or travel through several areas in sequence.

That makes the system more than a plate reader. It becomes a mobility analyzer, capable of stitching together location traces from a large camera network and then enriching those traces with personal data from other records.

Jay Stanley compared that kind of capability to a kind of dynamic geofence, except one that can expand and flex based on a user’s questions rather than a fixed boundary imposed in advance. In his view, that makes the surveillance more granular and potentially more intrusive than traditional location requests.

Stanley said the combination of wide-area cameras and AI-driven querying can go far beyond a static search box, making it possible to ask who was near someone repeatedly, where they traveled, and how often they returned.

Timeline of the Flock AI controversy

What happens next?

The immediate question is whether Flock will change the product before any broader deployment. Because the company says the tool is still being tested, the prompts and workflows described in the code may not be final. But the reporting suggests the underlying direction is clear: Flock is building a system that transforms camera data into a far more sophisticated investigative engine.

That will likely invite more scrutiny from civil-liberties groups, regulators, and lawmakers who are already wary of plate-reader networks. The next phase may focus less on whether the technology is powerful and more on whether it should exist in its current form at all.

For police departments, the attraction is obvious: faster searches, richer data, and fewer manual steps. For privacy advocates, the danger is equally obvious: a cheap and convenient way to convert routine movement into suspicion, and suspicion into identity.

Flock built its business on watching cars. OS Investigate suggests the company is now trying to teach police software how to watch people.