Flock Safety、警察向け次世代AIシステムコード公開へ

Aug 19, 2026 5:00 AM

Flock’s surveillance cameras have already sparked outrage. WIRED reconstructed its next-generation AI system, already in use by some police, to confirm it goes much further than tracking license plates.

Photo-Illustration: Jobanny Cabrera; Getty Images

Vehicle surveillance giant Flock Safety has told the public for years that its technology “cannot recognize, identify, or track individuals.” It has now built a system that does both, an artificial intelligence tool for police that can identify drivers and track vehicles by their patterns of movement alone, WIRED has learned.

Drawing on a network of cameras that logs the movements of drivers in more than 6,000 communities, the tool can pick out potential witnesses by how often their cars pass through a neighborhood, or surface a driver’s “associates” from the cameras they pass together. Because the system also reaches police case files, 911 dispatch logs, and commercial identity records, those plates can be turned into names, home addresses, and relatives. It can search for people in an area drawn on a map based on nothing more than a physical description.

The software, originally called Nightshift and more recently renamed OS Investigate, ships with 69 prewritten prompts that officers can select, review, or edit, and then submit to the AI. Officers can also input prompts of their own.

The suggested prompts sit in a cache of more than 450 files that WIRED found on Flock's own website, served by its login pages to anyone who loaded them. The code describes 45 tools at the AI’s disposal, giving it access to plate scans and camera metadata, arrest records, case files, dispatch logs, ballistics results, and commercial databases that contain Social Security numbers, dates of birth, phone numbers, email addresses, relatives and associates.

Flock says it is testing the product with a small group of law enforcement partners and describes it as still in development, with capabilities that may not reflect what it eventually sells. It arrives as the company faces bipartisan political pressure, a growing record of officers caught misusing its platform, and a wave of vandalism that has left cameras sawed off and lenses painted over in cities across the country.

Privacy and legal experts say Flock’s tools already raise serious constitutional concerns by giving police broad, often warrantless access to a vast record of people's movement. But the prompts WIRED examined go further, turning a tool for finding a known vehicle into one that casts suspicion on drivers based on where and how often they drive, regardless of whether they are wanted for a crime.

Flock built its business on a simple transaction. A camera photographs a passing car, converts the plate to text, and checks it against a list of vehicles police are looking for. Cars that are not on the list are recorded and otherwise ignored. OS Investigate’s pre-loaded prompts describe an inversion of that arrangement. An officer no longer needs a plate, a name, or a crime to begin. They supply a place, a stretch of time, and a pattern of behavior, and the system is designed to hand back the people who fit.

All 69 of the prompts are offered to officers as a menu of canned searches. Selecting one pastes it into a chat box, where the officer can review or edit it before clicking again to submit.

One prompt Flock pre-loaded into the system reads: “Find me witnesses based on vehicles most seen in [neighborhood] during [last 14 days] during [daily timeframe]. *(will not include whitelisted vehicles).” An officer would fill in the blanks and submit it. The output is a list of plates, which other tools in the product then convert into names and home addresses.

Another prompt instructs the system to list everyone arrested more than twice in two years for “any offense,” exempting only narcotics arrests, and then says to map where those people live, retrieve the calls for service at their homes, and “do a workup on the top 3 individuals.” The prompt begins with everyone in the area who has an arrest record and ends with dossiers on three of them, chosen by the software.

A “workup,” in Flock's terminology, is a one-command background check. It starts with a name and a date of birth and returns what the department's records and commercial data hold: vehicles, prior listings as a suspect, and, on a second screen, relatives, phone numbers, and online accounts.

Nineteen of the prompts describe hunting for patterns rather than looking up a record. Fourteen call for no plate, no name, and no description. The officer simply supplies a location, a window of time, and a behavior.

One group of requests asks for vehicles that visited three or more retail locations in a city in three days, or multiple banks in a week, or multiple gas stations between midnight and 5 am. None of the five names a target or an incident.

The tool behind those searches carries a filter that strips out buses, semi trucks, work vans, and trailers, so a delivery driver visiting five shops would come off the list, while an ordinary driver making the same rounds remains.

The same type of search also applies to travel between cities: vehicles that left one city for another and returned within a week, made repeat roundtrips over 14 days, or passed through three areas in sequence.

Flock did not respond to a list of questions about the product and did not dispute the capabilities described in our reporting. Instead, in a statement, spokesperson Paris Lewbel characterized OS Investigate as a separate product from Flock’s license-plate reader technology, designed to help investigators work across information their agencies already have access to. He said the tool’s capabilities and workflows may change significantly before a broader release.

Amid mounting controversy over how police use and abuse its cameras, Flock's long-held position has been that its technology records plates and not the drivers behind them. Its own public trust pages tell customers that its technology is built for specific, case-based investigations and “not for watching people.” Experts who reviewed the AI tool for WIRED say those assurances are hard to square with what they found in the code.

“I think that most people imagine that license plate readers are checking license plates and they're looking if a vehicle is wanted. And if it's not wanted, the police will move along,” says Jay Stanley, a senior policy analyst with the ACLU's Speech, Privacy, and Technology Project. “But we're increasingly seeing the aggressive deployment of AI by Flock. And in some ways, like China, very little space between what can be done and what is being done.”

One example, for instance, takes a single plate and asks the system to plot the journeys of that car and its “top 3 associates.” The code describes explicitly how the ranking works. It counts how often other plates appear at the same cameras within a two-minute window of the target by default, and reports back any car that shows up next to it three or more times within a “confidence threshold” defaulting to 0.75. An officer supplies one vehicle and the software finds up to 20.

| Got a Tip? |

| --- |

| Are you a current or former police officer, government official, or Flock Safety employee who wants to talk about the company? We’d like to hear from you. Using a nonwork phone or computer, contact the reporters securely on Signal at dmehro.89 and dell.3030. |

“I don't know how else to say this, but this sounds completely insane,” says Noel Pichardo, who was briefed on Flock’s technology during a pilot program while serving as a Pawtucket, Rhode Island, police officer and who reviewed the prompts for WIRED. “I don't know how anyone can argue against the idea that Flock literally tracks people.”

Pichardo spoke out against the technology before Rhode Island state lawmakers in 2024, weeks after he served a 30-day suspension for criticizing Flock and department leadership in a local newspaper. The state police chiefs' association publicly rebuked him days later. Investigations and discipline followed, which he believes were retaliatory, and he resigned as the department moved to fire him.

A detective at a California law enforcement agency, granted anonymity because he was not authorized to speak publicly, sees in OS Investigate a tool his department would use. Flock has closed many vehicle theft cases for them, he says, and officers have worked with similar technology for years. “I was not aware of this AI product, but it doesn't change the way I think about it,” he says. “We'll use whatever tool is available to solve crime. It isn't controversial.”

The detective said the witness-finding prompts made him slightly uncomfortable, but added, “As law enforcement, we need to use every tool available, in spite of what the ‘defund’ people say.”

WIRED analyzed the software by examining code Flock served from its own login portals. Loading those pages returned a user interface for screens that would ordinarily appear only after a police user signed in, letting reporters reconstruct portions of the authenticated applications, their prompts, search fields, tools, and controls, without actual access. The files appeared to contain everything needed to reconstruct the interface from scratch.

Before an officer runs a search, the software can ask for a justification. The code shows that by default an officer has to type a reason for the search; however, the form appears to place no requirements on what that reason says or how long it must be. If a department also requires a case number, the form checks only that it contains at least three characters. WIRED could not determine whether Flock applies additional checks on its servers after the information is submitted.

WIRED shared its findings with Buchodi, a pseudonymous independent security and privacy researcher who has spent more than a decade reverse engineering consumer software and surveillance technologies. Buchodi examined Flock's code separately and reproduced key parts of the analysis.

References to the tools Flock offers, the data they can apparently reach, and the suggested prompts Flock authored for police are fixed in the code WIRED reviewed. What that code cannot show is how the system behaves in use. It contains no instructions given to the model itself, and it does not reveal what Flock's servers do with a request once submitted, what they send back, or whether they refuse any command.

Flock has said little about the tool in public, but Garrett Langley, its chief executive, once likened it to a crime analyst who is, in his words, “always awake, always watching out for you.” At a demonstration for police chiefs in Denver last year, Langley described the ability to search every camera in a city for cars tailing an armored truck, pulling up the registered owners of those vehicles and checking their arrest records, in under a minute.

Investigators who try it, he said, get “addicted.”

The company Langley started in Atlanta, where his first camera was a phone in a waterproof box, now logs 20 billion plate scans a month. Andreessen Horowitz, Founders Fund, and other venture firms have poured money into the business, which was valued at $7.5 billion in its most recent funding round. The company’s original tagline was“the first public safety operating system that eliminates crime;” and Langley recently told Vanity Fair that Flock would have solved the disappearance of Nancy Guthrie, Savannah Guthrie’s mother, had its cameras been deployed where she vanished.

Flock counts roughly 140,000 monthly active users.. At the company’s new Atlanta headquarters, a gold-painted camera covered in signatures commemorates an earlier ritual of marking each additional $1 million in revenue with a new camera. “Now we do more like a million a day,” Langley recently told Vanity Fair.

Andrew Guthrie Ferguson, a law professor at George Washington University, says a system like this one was inevitable. “To use all the data collected, you need to build in the ability to query it,” he says. “To make things easier for police officers who are not going to create their own agentic tools, you must preprogram the system with some natural language prompts and automated workflows.”

The choice of prompts is what strikes him. “It is not rocket science to figure out [that] police will want to find certain objects or patterns, but it is interesting to see what they preprogrammed,” he says. “There are a lot of innocent activities that could be flagged, even if it also might be quite useful for investigators.”

Because the prompt menu is a convenience rather than a constraint, an officer can also type whatever he or she wants. Selecting a prompt drops its sentence into an ordinary chat box, where it can be edited, rewritten, or replaced outright.

Chad Marlow, an attorney who runs the ACLU's “Get The Flock Out” campaign, says that is where the danger lies. “The difference between a police officer saying, ‘Do you see any criminal patterns’ or ‘Find me criminal patterns’ is going to produce two different things,” he says. “And I think the fact that they don't seem to have any limitations on the prompts enables the police to almost lead the AI in what it does.”

Marlow says his worry is the officer who asks an AI which people it considers most likely to be committing crimes.“It's going to develop not a legal standard of criminal suspicion, but whatever that AI standard is.”

The code already describes the filters that question would run through. Flock's software chains vehicle searches to queries against police records that sort people by sex, race, ethnicity, height, weight, build, scars, marks, and tattoos. One prompt searches by description alone, asking for records matching “male, ~6ft, black hoodie, forearm tattoo” near a location. The tool allows an officer to bound those criteria by selecting a radius or drawing arbitrary shapes on a map.

Stanley says that closely resembles what the US Supreme Court took up in June, when it held that people keep a privacy interest in records of where they have been, and that the Fourth Amendment applies when police demand those records from a company. Here, police departments buy that capability outright, opening space for fishing expeditions on a scale he says has no precedent.

“You’ve got a literal geofence. But the AI can also bring it way beyond that kind of a static boundary, and who is inside that boundary,” he says. “Who was near this person for more than 10 minutes in the past month? Who was in these six locations more than three times? It can become much more fine-grained, sophisticated, flexible.”

Flock has been building the AI system while under scrutiny for how police have abused the cameras it already runs. A Texas officer searched more than 83,000 cameras across the country last year while looking for a woman who had a self-administered abortion, according to search logs first obtained by 404 Media. A recent Washington Post analysis found 50 officers charged with or accused of misusing plate reader systems, 26 of them to track wives, girlfriends, exes, and women they wanted to meet. Illinois regulators found last August that Flock had violated state law by running a now-paused pilot program that handed Customs and Border Protection access to cameras in the state.

Flock leaves most of the rules to its customers. Agency leaders decide who gets credentials, whether to share their camera data across city and state lines, and whether to switch on the settings the company offers to deter abuse. The audit tool that scans an agency’s own searches for suspicious patterns arrived in April 2026 as an option departments can enable. Many departments did little to monitor the searches themselves, the Post found.

The Electronic Frontier Foundation reviewed datasets pertaining to more than 12 million searches logged over 10 months and found that roughly 20 percent gave only a vague term like “investigation,” “suspect” or “query.” More than 50 agencies ran hundreds of searches whose reason fields referenced protest activity.

Flock says its newest safeguards will address some of that. The company announced last week that, by the end of the year, it will monitor for abnormal searches, lock out flagged users automatically, and require officers to attach case codes to their searches.

At the same time, the company is also hiring engineers to “drive the architectural evolution” of its new system, wire it into Flock’s larger data platform, and build a framework it says will act on behalf of investigators, according to two recent job postings. The role covers multi-step AI workflows and new capabilities including automated lead generation and cross-camera correlation—uses well beyond the version of the software WIRED examined.

“It is clear Flock has aspirations far beyond ALPRs to become a digital platform for policing,” Ferguson says. “We are on the cusp of the age of agentic policing and the ability to create natural language chatbots to search within massive datasets will soon become the norm.”

-

-

-

-

• Take our survey: Do you work in tech? We want to hear from you

Dhruv Mehrotra is an investigative data reporter at WIRED, where he uses technology to find, build, and analyze datasets for accountability journalism. His reporting has examined surveillance, policing, data brokers, platforms, and the government agencies and companies that use them.

Before joining WIRED, he worked at Bloomberg News, the Center for ... Read More

Dell Cameron is an investigative reporter from Texas covering privacy and national security. He's the recipient of multiple Society of Professional Journalists awards and is co-recipient of an Edward R. Murrow Award for Investigative Reporting. Previously, he was a senior reporter at Gizmodo and a staff writer for the Daily ... Read More

Read More