Skip to content

Build1 publisher3 min readPublished

Flock's OS Investigate starts the search before there is a suspect

WIRED found 14 prompts in Flock Safety's unreleased investigation tool that need no plate, name, or description. The predicate for the search becomes the buyer's problem.

The Engineer · Build desk

Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction

Photograph accompanying Flock's OS Investigate starts the search before there is a suspect
Photo: foxbusiness.com

What happened

  • WIRED reported on August 19 that Garrett Langley's next Flock Safety product, originally called Nightshift and now known as OS Investigate, can begin an investigation without a known license plate, person, or crime, letting officers search for people and vehicles through patterns of movement.
  • An officer supplies a place, a period, and a behavior, and OS Investigate returns people or vehicles that fit.
  • WIRED reported that 19 of the prompts focus on patterns, while 14 require no plate, name, or physical description.
  • Some OS Investigate prompts start with familiar police work: an officer can enter a plate and ask for associated vehicles seen near it.
  • Other prompts reverse the search: one asks for possible witnesses by finding vehicles most frequently seen in a neighborhood over the previous 14 days; another looks for vehicles visiting three or more retail locations in three days; similar searches target repeated visits to banks, overnight stops at gas stations, or round trips between cities.

Compiled by The EngineerSomething wrong?How this is made

Why it matters

WIRED reported on August 19 that Flock Safety's next product, originally called Nightshift and now known as OS Investigate, can begin an investigation without a known license plate, person, or crime [1]. That inverts the normal order of a query: instead of an identifier going looking for movement, an officer supplies a place, a period, and a behavior, and the system returns the people or vehicles that fit [2].

The prompt inventory is the story. Of the prompts WIRED counted, 19 focus on movement patterns and 14 require no plate, name, or physical description at all [3]. Some are recognisable police work, such as entering a plate and asking for vehicles seen near it [4]. Others run backwards: find possible witnesses by listing the vehicles seen most often in a neighbourhood over the previous 14 days, or find vehicles that visited three or more retail locations in three days, with variants for repeated bank visits, overnight stops at gas stations, and round trips between cities [5]. One prompt starts from people arrested more than twice in two years, maps their homes, pulls service calls at those addresses, and orders a detailed "workup" on three people the workflow itself selects [6].

Once a vehicle surfaces, the system can enrich the plate with names, addresses, relatives, phone numbers, online accounts, and prior appearances in police records [7], drawing on joins between vehicle scans, case files, arrest records, 911 dispatch logs, ballistics results, and commercial identity databases [8] assembled through a chat interface [9]. The intended payoff is speed [9].

What comes out are leads, not findings. An inferred association between two vehicles does not establish that their owners know each other, and repeated trips through a commercial district do not establish criminal conduct [10]. The specific hazard is presentational: a prewritten prompt makes an uncertain statistical relationship look like a routine database lookup [11].

Note how this was obtained. WIRED reconstructed parts of the interface from files exposed through Flock Safety's login pages, which show prompts, available tools, search forms, and intended workflows [12]. The files do not show the model's underlying instructions, the server-side checks applied to requests, which tools are live for current testers, or how the system behaves in operation [13]. Everything that would answer the compliance question is on the side that was not visible. Flock Safety told WIRED that OS Investigate is still in development with a small group of law enforcement partners, and spokesperson Paris Lewbel said it is separate from the company's licence plate reader product and that the observed capabilities may change before a wider release [14].

That leaves the predicate with the purchaser. If the tool will generate candidates from behaviour alone, the agency that buys it owns the policy on who may run a no-identifier search, what justification gets logged, and how a statistical hit is described later in an affidavit. Flock has the capital to keep shipping into that gap: it has already expanded from fixed plate readers into video cameras, gunshot detection, drones, 911 software, and real-time crime centre tools [15], and its March 2025 round raised $275 million at a $7.5 billion valuation led by Andreessen Horowitz [16], roughly eight years after the company was founded in 2017 [17], with proceeds earmarked for product innovation, research and development, and US-based manufacturing [18].

Watch whether the 14 no-identifier prompts survive to general availability, and whether Flock ships a required justification field and audit log alongside them. Watch the first non-pilot procurement documents for language on who may run pattern searches. And watch whether the pilot agencies are ever named, since the small partner group is currently the only place this behaviour is being tested [14].

Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories