This Is Flock’s AI Search Tool for Cops

WIRED has reverse-engineered Flock Safety’s newest AI-powered search interface, which ships as browser code to police officers. The tool lets law enforcement type a natural-language description—"man in a red hoodie"—and have the system scan across multiple surveillance cameras to surface matching individuals. This is a significant leap from Flock's earlier license-plate recognition, moving toward general person-tracking based on appearance. The analysis reveals both the technical maturity of real-time visual search and the operational risks of coupling large-scale camera networks with AI that can misinterpret or amplify biases. Flock’s pivot from license-plate readers to full-fledged person search is a watershed moment for civilian surveillance. The company has already deployed thousands of cameras across US neighborhoods, and now it’s layering an AI that can track a person across a city based on a verbal description. As WIRED’s rebuild shows, this isn’t a distant prototype—it’s live code running in police browsers today. The technical feat is impressive: translating a sentence into a multi-camera cross-reference of appearance features in near real time. But the implications are deeply unsettling. The core issue is scale and permanence. License plates are a proxy for vehicle identity, but they’re not biometric. Person search based on appearance—clothing, build, gait—becomes a de facto biometric tracking system, especially when combined with the temporal and spatial data from Flock’s network. Even if the AI is only “suggesting” matches, the sheer volume of false positives in a diverse population can lead to wrongful stops, especially for people of color. The WIRED investigation highlights how the tool’s interface presents matches without confidence scores, nudging officers to treat guesses as leads. What’s striking is the asymmetry of power: residents have no way to know they’re being indexed by this system, let alone to opt out. Flock’s marketing frames it as a “public safety” tool, but the underlying architecture is a surveillance infrastructure that grows more capable with each software update. The fact that the code is delivered to a browser—not a hardened server—suggests rapid iteration and a lack of external oversight. This is a classic “move fast and break things” approach applied to constitutional rights. The broader trend here is the commodification of surveillance. Flock is not a government agency; it’s a private company selling a service, and its incentives are to expand usage and feature set. The WIRED rebuild is a valuable public service, showing exactly what the tool does and doesn’t do. It should force a conversation about whether we want this kind of AI in the hands of every local police force, and what safeguards—if any—are acceptable. The answer isn’t to ban AI, but to demand transparency, auditability, and a legal framework that prevents this from becoming a permanent dragnet. Source: https://www.wired.com/story/flock-ai-search-user-interface/