Flock Safety just got caught building something it never fully explained to the public: an AI-powered search tool that lets police describe a person in plain English and have algorithms hunt for them across an entire network of surveillance cameras. WIRED didn't wait for Flock to explain it. Reporters rebuilt the tool themselves from the code the company quietly ships to officers' browsers, and what they found raises fresh questions about how far automated policing has already gone.
It turns out one of the most consequential AI products in American policing didn't come with a press release. It came wrapped inside browser code, sitting quietly on the client side of a police department's dashboard, waiting for someone to go looking for it. That's exactly what reporters at WIRED did, pulling apart the interface that Flock Safety sends to officers and rebuilding it piece by piece to understand what the company's newest AI search tool actually does. The answer: it lets a cop type in a plain-language description, say, a red hoodie and a silver sedan, and have an AI system scan across multiple camera feeds at once looking for anything that fits.
That's a meaningful jump from what most people picture when they hear "license plate reader," which is the category Flock built its business on. The company has spent years wiring up cities with automated cameras that log vehicle plates, feeding a searchable database used by thousands of police departments nationwide. Layering a generative AI search function on top changes the nature of the tool entirely. Instead of matching an exact plate number, officers can now describe a person or vehicle in natural language and let a model do the pattern-matching across camera networks, effectively watching in multiple places at once for someone who fits a loose description.
WIRED's investigation matters because none of this was clearly disclosed by Flock in a way the public, or even many of the departments deploying it, could easily audit. Instead, the functionality lived inside code shipped straight to an officer's browser, the kind of implementation detail that normal product marketing skips right over. Reporters Dell Cameron and Dhruv Mehrotra had to essentially do the reverse engineering work themselves, treating the browser as the primary source document since the company hadn't laid out the tool's mechanics in any public-facing way.
This fits a pattern that's become familiar in the surveillance tech world. Companies roll out increasingly powerful computer vision and pattern-matching tools to police customers first, often with limited public documentation, and details only surface once journalists or researchers dig into the actual code or file public records requests. Flock has faced this kind of scrutiny before over how its camera network operates and how broadly data gets shared across jurisdictions, but an AI system capable of searching for a written description across cameras represents a meaningfully different capability than plate matching, since it introduces the kind of subjective, description-based matching that has historically been prone to bias and error in facial recognition and similar systems.
Civil liberties groups have long warned that expanding automated surveillance without clear guardrails risks entrenching mistakes at scale, especially when a written description, which can be vague, incomplete, or simply wrong, becomes the input an algorithm uses to flag potential suspects across an entire city's camera grid. There's no indication in WIRED's reporting that Flock has published accuracy rates or bias testing for this specific tool, which is exactly the kind of transparency gap that tends to draw attention from lawmakers once a story like this breaks.
What happens next probably depends on how loudly this reporting travels. Flock has built a business model around selling its technology directly to local police departments, city by city, often without much state or federal oversight standing in the way. If this AI search capability gets adopted as widely as license plate reading has, the debate over consent, accuracy, and civil liberties that has surrounded automated license plate readers for years is likely to resurface, just aimed at a more powerful and less understood tool. Watch for public records requests, city council pushback, and possibly regulatory inquiries as more departments start explaining, or get asked to explain, exactly what this AI system can see and how confident it actually is when it flags a match.
For readers who've assumed AI surveillance tools come with clear public disclosure before they hit the streets, this story is a reminder that a lot of this technology gets built quietly and explained only after someone bothers to take it apart. Flock's expansion from plate-matching into description-based AI search shows how fast the capabilities of everyday policing tools can outpace the public conversation about whether, and how, they should be used.
