Vehicle surveillance technology manufacturer Flock Safety has asserted publicly to communities and municipal leaders for years that its license plate reader technology is strictly limited to vehicle identification and cannot recognize, identify, or track individual human beings. However, extensive technical disclosures reveal that the company has engineered an advanced artificial intelligence surveillance platform capable of achieving both functions. The newly developed law enforcement tool can identify individual drivers and track motor vehicles solely by analyzing their historical patterns of movement across public roads, relying on an expansive nationwide network of automated cameras installed in thousands of neighborhoods.
Drawing on an interconnected surveillance grid of optical cameras that logs the movements of drivers across more than 6,000 communities nationwide, the software can pinpoint potential witnesses based on how frequently their cars travel through a designated neighborhood. The platform is also engineered to surface a targeted driver's close associates by evaluating which secondary vehicles routinely pass the same camera locations within narrow timeframes. Because the AI software integrates directly into municipal police case databases, emergency 911 dispatch logs, and commercial identity records, captured license plate numbers can be instantly converted into full legal names, verified residential street addresses, personal phone numbers, and listed family relatives. Furthermore, police officers can execute targeted geographic searches across map interfaces based on nothing more than basic physical human descriptions.
The software platform, originally designated internally under the project name Nightshift and more recently rebranded as OS Investigate, ships to law enforcement customers with a pre-configured menu of 69 suggested natural language prompts. Operating the interface, police officers can select from these pre-written queries, modify specific search parameters, or submit original custom prompts directly to the artificial intelligence engine. These capabilities represent a dramatic departure from passive license plate monitoring, transforming automated cameras into an active, predictive surveillance engine capable of analyzing population-wide movements.
The Exposed Codebase: 69 Pre-Loaded Prompts and 45 Autonomous Tools
The collection of pre-loaded operational prompts was discovered within a public cache of more than 450 code files hosted on Flock Safety's official web portal. These files were served directly by the platform's login pages to any web browser loading the interface, granting unauthenticated access to the underlying front-end codebase. Inspection of the source code reveals 45 specialized internal tools at the AI agent's disposal. These automated tools grant the software direct access to historical license plate scans, camera metadata, municipal arrest records, active police case files, 911 emergency dispatch logs, ballistics laboratory results, and commercial public records databases containing Social Security numbers, dates of birth, telephone numbers, email addresses, and detailed family and associate listings.
Flock Safety has stated publicly that it is currently testing the software product with a limited group of law enforcement partners. The vendor emphasizes that the tool remains under active development, noting that features present in the current codebase may not accurately reflect the finalized software version offered for commercial sale. The disclosure of this AI technology arrives as Flock Safety faces mounting bipartisan political pressure, a growing record of documented police officer misuse across multiple states, and a nationwide wave of physical vandalism that has left municipal surveillance cameras sawed off mounting poles or sprayed over with paint in numerous cities.
Legal scholars and civil liberties advocates contend that Flock Safety's existing network of license plate readers already poses profound Fourth Amendment constitutional concerns by providing police officers with broad, warrantless access to the detailed movement histories of millions of citizens. However, the operational prompts examined in the codebase push surveillance capabilities far beyond traditional hotlist checks. Rather than functioning as a tool to locate a specific vehicle known to be involved in a crime, OS Investigate casts broad suspicion on drivers based entirely on where and how frequently they travel, regardless of whether they are suspected of any unlawful activity.
Historically, Flock Safety established its core enterprise business model around a simple, single-purpose transaction. An automated camera photographs a passing motor vehicle, converts the license plate characters to text using optical character recognition, and checks the number against a hotlist of stolen or wanted vehicles provided by police agencies. Vehicles not appearing on a hotlist are recorded into a database and otherwise ignored. The pre-loaded prompts embedded within OS Investigate represent a complete inversion of that original arrangement. Under the new AI paradigm, an investigating officer no longer requires a license plate string, a suspect's legal name, or an existing crime report to initiate a search. Instead, the officer provides a geographical area, a timeframe, and an observed pattern of behavior, and the AI system is engineered to hand back a list of matching individuals and their personal identities.
Detailed Breakdown of Canned Prompts and Automated Background Check Workups
All 69 pre-programmed prompts are presented to police officers as a convenient menu of canned investigative searches. Selecting a prompt automatically pastes the text into an interactive chat interface, where the officer can review, edit, or customize the wording before clicking to submit the query to the processing engine. This natural language design enables officers to conduct complex database cross-referencing without requiring specialized technical training.
One representative prompt pre-loaded into the software by Flock Safety reads: "Find me witnesses based on vehicles most seen in [neighborhood] during [last 14 days] during [daily timeframe]. *(will not include whitelisted vehicles)." To execute the search, an officer fills in the bracketed variables specifying the location and time window and submits the query. The AI system processes the request and outputs a list of license plates. Integrated identity modules within the product then automatically convert those license plate strings into full legal names, home addresses, and personal contact profiles.
Another pre-configured prompt instructs the system to analyze local criminal records and assemble a list of every individual arrested more than twice within a two-year span for "any offense," explicitly exempting only narcotics-related arrests. The prompt then directs the software to map the home addresses of those individuals, retrieve all historical 911 calls for service logged at those residences, and "do a workup on the top 3 individuals." Through this single instruction, the software initiates a query with everyone in an area holding an arrest record and concludes by generating comprehensive intelligence dossiers on three individuals selected automatically by the algorithm.
In Flock Safety's technical vocabulary, a "workup" is an automated, single-command background check. The process begins with an individual's name and date of birth and queries both municipal law enforcement records and commercial data aggregators. On the primary screen, the software displays registered motor vehicles and prior police listings where the person was identified as a suspect. On a secondary interface screen, the system compiles known relatives, associated telephone numbers, and linked online accounts.
Pattern Recognition Tracking: Searching Behavior Without License Plates or Suspect Names
A detailed review of the prompt menu indicates that 19 of the pre-loaded queries focus on searching for behavioral patterns rather than looking up specific law enforcement records. More significantly, 14 of the prompts require no license plate number, no suspect name, and no physical person description whatsoever. The investigating officer simply supplies a geographic location, a temporal window, and a specified pattern of physical movement across the road network.
One cluster of pre-configured requests directs the AI engine to identify motor vehicles that visited three or more commercial retail locations within a city over a three-day period, multiple banking facilities within a single week, or multiple gas stations between the hours of midnight and 5:00 AM. None of these five query templates require a target suspect name or an underlying incident report to execute.
To refine search results and eliminate routine commercial traffic, the underlying search tools incorporate an analytical filter that automatically strips out public transit buses, commercial semi trucks, utility work vans, and cargo trailers from the output. Consequently, a commercial delivery driver visiting five retail stores on a delivery route is automatically removed from the search results, whereas an ordinary citizen making identical trips across town remains flagged on the police review list.
The same pattern-recognition methodology applies to inter-city travel tracking. Prompts enable officers to isolate vehicles that departed one city for another and returned within seven days, conducted repeat roundtrips over a 14-day window, or passed through three designated geographic areas in fixed chronological sequence.
When questioned regarding the product capabilities, Flock Safety spokesperson Paris Lewbel stated that OS Investigate is a separate product platform from the company's core license plate reader technology. He explained that the system is designed to assist investigators in working across disparate information sources that their agencies already possess legal authority to access. Lewbel added that the tool's features and operational workflows remain under development and may undergo significant changes before any widespread commercial release.
Co-Location Algorithms: Ranking Associates and Mapping Social Networks
Despite mounting controversy over how law enforcement agencies deploy surveillance hardware, Flock Safety has consistently maintained publicly that its technology records vehicle license plates rather than individual human drivers. The vendor's public trust web pages explicitly assure clients and citizens that its systems are built for specific, case-bound investigations and are "not for watching people." However, privacy and legal scholars who analyzed the exposed AI codebase emphasize that these public assurances are fundamentally incompatible with the automated capabilities built into OS Investigate.
Jay Stanley, a senior policy analyst with the ACLU Speech, Privacy, and Technology Project, observed that the public generally understands license plate readers as simple tools that check passing cars against hotlists of stolen or wanted vehicles. However, the aggressive deployment of artificial intelligence by vendors like Flock Safety narrows the gap between technical capability and active deployment, establishing pervasive surveillance structures reminiscent of authoritarian models seen in countries like China.
The technical code reveals the exact mathematical formula governing the software's co-location search tool. Given a single target license plate, the software can plot the historical travel routes of that car along with its "top 3 associates." The algorithm calculates associate rankings by counting how frequently secondary license plates appear at the same camera locations within a default two-minute window of the primary target vehicle. If a secondary vehicle appears alongside the target three or more times within a default confidence threshold of 0.75, it is reported back to the officer. Through this automated correlation, a query commencing with a single vehicle allows the software to identify up to 20 associated vehicles and their owners.
Former Pawtucket, Rhode Island police officer Noel Pichardo, who was briefed on Flock Safety's technology during a departmental pilot program and subsequently reviewed the prompt codebase, expressed deep alarm over the system's capabilities. Pichardo testified against the expansion of surveillance technology before Rhode Island state lawmakers in 2024, weeks after serving a 30-day suspension for criticizing vendor policies and police leadership in a local publication. Following subsequent internal investigations and disciplinary actions that he characterized as retaliatory, Pichardo resigned from the police force as the department initiated termination proceedings. The state police chiefs' association also issued a public rebuke against him.
In contrast, an active detective serving in a California law enforcement agency indicated that OS Investigate represents an invaluable tool that police departments would eagerly adopt. Noting that Flock Safety cameras had solved numerous vehicle theft cases for his department, he argued that law enforcement officers must utilize every available technology to solve crimes. While acknowledging that witness-finding prompts made him slightly uncomfortable, he emphasized that police agencies must deploy advanced tools regardless of opposition from civil rights advocates.
Constitutional Rights and the Expanding Horizon of Warrantless Mass Surveillance
Examination of the front-end user interface revealed the technical mechanics of search justification forms. Before an officer can run a query, the software prompts the user to enter an investigative rationale. However, the code demonstrates that the default input field places no minimum length requirements or content validation checks on what the officer writes. If a police department mandates a case file number, the system verifies only that the entry contains at least three characters. Whether Flock Safety enforces secondary validation checks on its internal servers after data submission remains unconfirmed in the front-end code.
Independent security and privacy researcher Buchodi, who possesses over a decade of experience reverse-engineering consumer software and surveillance technologies, separately examined Flock Safety's exposed web assets. Buchodi successfully reproduced key aspects of the technical analysis, confirming the presence of the pre-loaded prompts, tool definitions, and system controls in the public code files.
While the front-end code establishes the tools and prompt libraries created for law enforcement, it does not reveal the underlying system instructions provided directly to the AI model itself, nor does it disclose how Flock Safety's backend servers process, filter, or execute queries once submitted by an officer.
Flock Safety Chief Executive Officer Garrett Langley has publicly characterized the AI platform as an automated crime analyst that remains continuously active and vigilant. At a demonstration conducted for police chiefs in Denver, Langley showcased the software's capability to search every camera across a municipality for vehicles tailing an armored truck, retrieve registered owner identities, and cross-reference criminal arrest records in under one minute. Langley noted that law enforcement investigators who test the technology quickly become dependent on its speed and scope.
From Atlanta Startup to $7.5 Billion Giant: Commercial Expansion and Political Scrutiny
Founded in Atlanta, Flock Safety began operations with a single prototype camera housed inside a waterproof enclosure. Today, the company's surveillance network logs approximately 20 billion license plate scans per month across the United States. Prominent Silicon Valley venture capital firms, including Andreessen Horowitz and Founders Fund, have invested heavily in the business, driving its valuation to $7.5 billion in its most recent funding round. The vendor's original corporate tagline declared itself "the first public safety operating system that eliminates crime." In a recent interview, Langley asserted that Flock Safety's camera grid would have solved the high-profile disappearance of Nancy Guthrie, mother of television host Savannah Guthrie, had cameras been installed where she vanished.
Flock Safety currently reports approximately 140,000 monthly active users across law enforcement agencies. At its new corporate headquarters in Atlanta, a gold-painted surveillance camera covered with employee signatures commemorates an early company tradition of celebrating every $1 million in revenue with a new camera installation. Langley noted that the enterprise now generates approximately $1 million in revenue per day.
George Washington University Law Professor Andrew Guthrie Ferguson noted that the emergence of AI-driven natural language querying was inevitable given the massive volume of surveillance data collected by private vendors. He explained that pre-programmed prompts and automated workflows allow officers without technical programming skills to execute complex queries, but warned that such systems routinely risk flagging entirely lawful civilian behavior for criminal investigation.
Chad Marlow, an attorney managing the ACLU's "Get The Flock Out" campaign, highlighted the danger of unconstrained natural language prompt fields. He noted that variations in how an officer phrases a query can alter the AI's output, potentially leading the system to generate suspicion based on arbitrary algorithmic thresholds rather than established legal standards of reasonable suspicion.
The software codebase reveals that vehicle movement queries can be chained to law enforcement records filtering individuals by demographic attributes, including gender, race, ethnicity, height, weight, body build, scars, marks, and tattoos. One pre-loaded prompt searches specifically for individuals matching a description of "male, ~6ft, black hoodie, forearm tattoo" near a designated location. Officers can bound these searches by selecting a radius or drawing custom geometric shapes directly onto map interfaces.
ACLU analyst Jay Stanley pointed out that these AI geofencing capabilities closely align with issues evaluated by the US Supreme Court in its landmark June ruling regarding Fourth Amendment privacy protections for location data. He warned that AI tools allow police to extend location tracking far beyond static boundaries, enabling automated queries that track who was near a specific person or who visited multiple designated locations over extended time periods.
A History of Platform Misuse and the Pursuit of Technical Safeguards
Flock Safety's development of agentic AI technology comes amidst sustained scrutiny over documented instances of officer misuse across its existing camera network. According to search logs obtained during a privacy investigation, a Texas police officer executed searches across more than 83,000 cameras nationwide while attempting to track down a woman who had undergone a self-administered abortion. A separate media analysis of law enforcement disciplinary records identified 50 police officers charged with or accused of misusing license plate reader systems, with 26 cases involving the unauthorized tracking of spouses, girlfriends, former romantic partners, and women they sought to pursue.
In August, regulatory authorities in Illinois determined that Flock Safety had violated state law by running an unauthorized pilot program that provided federal Customs and Border Protection agents with access to camera networks across the state. That program was subsequently paused following state regulatory intervention.
An audit of search datasets conducted by the Electronic Frontier Foundation, analyzing more than 12 million logged queries across a 10-month timeframe, revealed that approximately 20 percent of search entries contained vague justifications such as "investigation," "suspect," or "query." Furthermore, the study discovered that more than 50 law enforcement agencies logged hundreds of searches explicitly referencing political protest activity.
In response to mounting criticism, Flock Safety announced planned security enhancements scheduled for implementation by the end of the year. The company stated that its systems will monitor for abnormal search patterns, automatically lock out flagged user accounts, and mandate that officers attach valid case file codes to every search request. Simultaneously, recent corporate job postings reveal that Flock Safety is actively hiring software engineers to expand the architecture of its new system, wiring it into larger data platforms to build multi-step AI workflows, automated lead generation modules, and cross-camera correlation capabilities that far exceed the software version revealed in the codebase.
As Professor Ferguson summarized, law enforcement is entering the era of agentic policing, where natural language chatbots querying massive surveillance databases will rapidly become standard operational procedure for police departments nationwide.
