llms.blog
AI

Leaked Flock Safety Code Exposes OS Investigate AI System for Police Surveillance

A technical analysis of client-side code exposed on Flock Safety's login portals has revealed OS Investigate, an unannounced artificial intelligence platform designed to track individuals and analyze vehicular travel patterns across police departments nationwide. The findings, first reported by WIRED and verified by independent security researchers, detail an AI-driven investigative system that links automated license plate reader (ALPR) networks with police databases and commercial records. Fl

axc
axc
2 min
Leaked Flock Safety Code Exposes OS Investigate AI System for Police Surveillance

A technical analysis of client-side code exposed on Flock Safety's login portals has revealed OS Investigate, an unannounced artificial intelligence platform designed to track individuals and analyze vehicular travel patterns across police departments nationwide. The findings, first reported by WIRED and verified by independent security researchers, detail an AI-driven investigative system that links automated license plate reader (ALPR) networks with police databases and commercial records.

Flock Safety's surveillance infrastructure currently operates across more than 6,000 communities in the United States. While the company has historically stated that its cameras only capture vehicle license plates rather than tracking individual people, the reconstructed codebase demonstrates direct capabilities for individual identification, behavioral profiling, and associate mapping.

Pre-Configured AI Prompts and Data Tooling

The exposed application files detail 45 specialized data tools and 69 pre-written AI prompts available to law enforcement users. Rather than querying single license plates associated with active warrants, OS Investigate enables open-ended behavioral queries that cross-reference:

  • Plate read timestamps and camera geographic coordinates
  • 911 dispatch calls, arrest records, and incident case files
  • Integrated ballistics test results
  • Commercial identity databases containing Social Security numbers, dates of birth, telephone numbers, known addresses, and family relations
Flock Safety OS Investigate Data Ingestion Pipeline

Behavioral Pattern Queries and Automated Dossiers

The system architecture marks an inversion of traditional license plate reader workflows. Instead of checking observed plates against static hotlists, the AI tool allows investigators to query movement patterns without specifying a suspect, vehicle plate, or active crime.

Pre-loaded prompts in the codebase include:

  • Witness and Frequency Discovery: Automated searches identifying vehicles most frequently recorded in a specific neighborhood during specified time windows, excluding whitelisted residents.
  • Associate Mapping: Algorithms that identify co-traveling vehicles by calculating camera pass timestamps within two-minute windows, surfacing top associates based on a 0.75 confidence threshold.
  • Pattern-of-Life Tracking: Filters detecting vehicles visiting multiple retail stores, banks, or gas stations across distinct municipal jurisdictions within defined multi-day windows.
  • Automated Individual Workups: One-command dossier generation combining local arrest histories, calls for service, and commercial identity records to rank and profile top individuals in a designated geographic area.

Flock spokesperson Paris Lewbel confirmed that OS Investigate is currently in pilot testing with select law enforcement partners, stating that features and workflows remain subject to revision before any broader commercial deployment.

Sources

AISurveillancePolicyComputer VisionSecurity
ShareX (Twitter)LinkedIn

Written by

axc

More to read

  • Sharpness-Aware Minimization in Large Language Models: How Adversarial Weight Perturbations and Flat Minima Boost Generalization

    In overparameterized deep neural networks, minimizing empirical training loss is insufficient to guarantee optimal generalization on unseen distributions. Modern deep architectures, including vision models and autoregressive Large Language Models (LLMs), operate in regimes where parameter counts far exceed training token counts, producing highly non-convex loss surfaces populated by infinite global minima. Standard optimization via Stochastic Gradient Descent (SGD) or AdamW often converges to sh

    1 min

  • Prefix-Tree KV Cache Management in Production: RadixAttention, Tree-Structured LRU Eviction, and Token-Level Sharing in SGLang and vLLM

    Prefix-Tree KV Cache Management in Production: RadixAttention, Tree-Structured LRU Eviction, and Token-Level Sharing in SGLang and vLLM Autoregressive large language model inference is heavily constrained by memory bandwidth and the computational overhead of the prefill phase. For workloads such as multi-turn conversations, autonomous agent tool loops, few-shot prompt evaluations, and tree-search decoding, consecutive requests often share substantial token prefixes. In a standard multi-turn ses

    1 min

  • Neural Collapse: How Simplex Equiangular Tight Frames Emerge at the Terminal Phase of Training

    In classification tasks, deep neural networks exhibit an unexpected geometric simplicity during late-stage optimization. While the internal activations of early training appear high-dimensional and complex, the penultimate layer representations and linear classifiers converge toward an exact, symmetrical geometric structure known as Neural Collapse (NC). First identified empirically by Papyan, Han, and Donoho (2020), Neural Collapse emerges during the Terminal Phase of Training (TPT). This regi

    1 min