In brief
- Flock Safety's OS Investigate ships with 69 prewritten prompts that give police AI access to plate scans, arrest records, and commercial databases holding Social Security numbers and relatives.
- Wired reports that 41 of the prompts require no special permission, and 14 need no plate, name, or description— just a place, a time, and a pattern of behavior.
- One prompt finds a car's "associates" by how often other vehicles appear at the same cameras within two minutes, flipping Flock's promise that it doesn't track people.
Flock Safety, the surveillance company whose cameras log vehicles across more than 6,000 U.S. communities, has built an AI tool that can identify and follow drivers using nothing but their movement.
The findings come in a report from Wired after it pulled the system's code from files on Flock's own site served to anyone who loaded its login pages.
The software—first called Nightshift, now renamed OS Investigate—is still in development and in testing with a small group of law enforcement partners, the company said.
An officer opens a chat box and picks from a menu of 69 canned prompts, or writes their own. One built-in query lists "witnesses" as the vehicles most seen in a neighborhood over the past two weeks.
Another finds everyone arrested more than twice in two years—exempting only drug offenses—then builds dossiers on the top three. A "workup" turns a name and birth date into relatives, phone numbers, and online accounts pulled from police records and commercial data brokers.
Fourteen of the prompts need no plate, name, or description at all. The officer supplies a location, a window of time, and a behavior—cars that hit three retail stores in three days, or several banks in a week—and the system returns the people who fit. A filter strips out buses and delivery vans, so an ordinary driver making those rounds stays on the list.
The associate-finding logic is spelled out in the code. It counts how often other plates appear at the same cameras within a two-minute window of a target car, flags any that show up three or more times above a 0.75 confidence score, and hands back up to 20 names.
"I don't know how else to say this, but this sounds completely insane," Noel Pichardo, a former Pawtucket, Rhode Island, police officer who reviewed the prompts, told Wired. "I don't know how anyone can argue against the idea that Flock literally tracks people."
Jay Stanley, a senior policy analyst at the ACLU, said the gap between what Flock's cameras can do and what the company is already doing leaves "very little space... like China."
Flock has long told the public its technology "cannot recognize, identify, or track individuals" and is "not for watching people." Spokesperson Paris Lewbel called OS Investigate a separate product from the company's license-plate readers, built to help investigators work across records their agencies already hold, and said its capabilities may change significantly before any wider release. The company didn't dispute Wired's reporting.
The confrontation has already spilled into the streets. In June and July, activists spray-painted, taped over, and destroyed Flock cameras across the country; The Guardian counted at least 33 incidents across 23 states. The anger tracks with how police have actually used the network. A report by the Institute for Justice found over two dozen cases of officers resigning or being arrested for allegedly using license-plate readers to stalk current or former partners.
The pushback has reached Capitol Hill, where Rep. Thomas Massie's planned bill would withhold federal funding from agencies that deploy the technology.
Researchers have built AI-generated patterns that confuse the plate-reading software, and Flock isn't alone in pushing the boundary: Meta has patented cameras that recognize faces and log a person's actions.