Flock's New Police AI Builds Social Graphs of Drivers

WIRED obtained the source code and a cache of more than 450 supporting files for Flock Safety's new AI investigative assistant, originally called Nightshift and now OS Investigate. The product draws on Flock's network of license-plate-reading cameras, which now span more than 6,000 communities.

The system ships with 69 prewritten prompts that an officer can select, edit, or replace with a custom query. Underneath those prompts sit 45 internal "tools" that give the AI access to plate scans, camera metadata, arrest records, case files, 911 dispatch logs, ballistics results, and commercial identity databases containing Social Security numbers, dates of birth, phone numbers, email addresses, and lists of relatives and other drivers the cameras have repeatedly seen near them. WIRED found the prompt and tool cache served by Flock's own login pages, accessible to anyone who loaded the URLs.

Three of the capabilities in the code go beyond the longstanding license-plate pitch. The first is a social graph of drivers built from co-travel: when a plate is read at multiple cameras, the system can list other vehicles that pass the same cameras around the same time, and rank them as the original driver's "associates." The second is a witness-surfacing prompt that looks at how often a vehicle passes through a neighborhood. The more often a car shows up, the higher it floats on a "potential witness" list, regardless of any wanted status. A third capability in the code lets an officer draw a shape on a map and ask the system to identify people inside it from a physical description alone, with no plate needed to start.

For years, Flock's public posture has been that the technology "cannot recognize, identify, or track individuals." That sentence lives in marketing copy and in replies to city council questions. On their own, the cameras do not perform facial recognition, and a plate read does not return a name. The code WIRED obtained describes a different system. It turns movement data into names, addresses, families, and a running list of who shows up where.

Flock says OS Investigate is in testing with a small group of law enforcement partners and is still in development, with capabilities that may not reflect what it eventually sells. The company did not address WIRED's specific findings about the social graph or the map-description search. The file review shows what the system is built to do, not what every officer is doing today. The constitutional question, privacy and legal experts quoted in the piece note, is that a warrantless query against movement data and a commercial identity database is itself a search, and that layering an AI on top of that pipeline changes the scale even if it does not change the legal theory. For residents whose plates feed a network of 6,000 communities, the difference between "the cameras read plates" and "the cameras build a social graph" is the difference between a log entry and a profile.

Flock faces bipartisan pressure, a growing record of officers caught misusing its platform, and a wave of vandalism, including sawed-off cameras and painted-over lenses, in multiple cities. State and federal litigation over warrantless access to ALPR (automatic license plate reader) data is already moving. The OS Investigate cache lands in the middle of that fight, and it gives city council members, police chiefs, and state legislators something they did not have a week ago: a falsifiable, document-based picture of what the vendor's own code says the tool can do.

The 450 files on Flock's own website, served by its own login pages, are the next fact-check target. If the file list and the code match what Flock says it sells, the public "cannot identify" line needs a rewrite.