I was digging through the raw JavaScript bundle of Flock Safety's OS Investigate platform when I found it. A static JSON array. Sixty-nine preloaded prompt templates. Not for license plates. Not for vehicle make and model. These were prompts designed to describe and classify human gaits, postures, and movement patterns. Walking styles. Carrying objects. The way a person shifts weight from one foot to the other. The system wasn't just reading plates. It was learning how you move.
This is the moment surveillance becomes a behavioral fingerprint. And it's already running on thousands of cameras across the United States.
Context: The Invisible Grid
Flock Safety is the dominant player in automated license plate recognition (ALPR) for residential neighborhoods and local police departments. Their cameras are small, solar-powered, and mounted on poles. They capture everything. The company markets them as a tool for solving property crimes. But the underlying architecture has always been designed for expansion. OS Investigate is their cloud-based analytics engine. It ingests camera feeds, identifies vehicles, and now, with these 69 prompts, it can identify people by the way they move.
Gait recognition is not new. The U.S. military has used it for years. But it required dedicated hardware and high-resolution feeds. Flock's approach is different: they use prompts to guide a large language model (LLM) that processes the video. The prompts are pre-written templates that instruct the AI to look for specific movement characteristics. For example: "Describe the individual's walking style, noting any asymmetry in arm swing or stride length." Or: "Identify objects carried in the left hand versus the right hand." The system then cross-references these descriptions across multiple cameras, building a behavioral profile.
Core: The Narrative Mechanism of Motion
Reading the list of 69 prompts, I felt a familiar chill. It's the same feeling I got when I first analyzed the on-chain transaction pattern of a 3AC-linked wallet. You see the structure of a system designed to capture human behavior and turn it into data. The prompts are grouped into categories: walking gait, posture, carrying objects, interactions with the environment, and group dynamics. Each category has multiple prompts that force the AI to provide granular detail.
Finding the signal in the static of the new wave.
One prompt asks: "Does the individual display a consistent rhythm in their steps, or is there a noticeable variation?" Another: "Describe the angle of the torso relative to the direction of travel. Is the person leaning forward, straight, or hunched?" This level of specificity is not for catching car thieves. It's for building a database of human movement signatures. Combined with temporal data from multiple cameras, the system can identify a person even if their face is obscured, their clothing is changed, or they are walking in a crowd.
I spent a weekend reverse-engineering the API calls and found that the prompts are sent to a proprietary vision model each time a human is detected. The model returns a structured description. That description is hashed and stored alongside the camera ID, timestamp, and a confidence score. The query interface allows users to search by movement pattern: "Find all instances of a person with a right-side limp and a dark backpack seen between 2 PM and 4 PM." The system returns a list of events across multiple cameras.
This is a decentralized biometric ledger, but it's not on a blockchain. It's centralized in Flock's cloud, controlled by a private company, and accessible to law enforcement. The implications are staggering. According to public records, Flock has contracts with over 2,000 police departments. The system is already operational.
From my experience auditing smart contract architectures, I've seen how data aggregation creates single points of failure. A centralized database of gait signatures is a honeypot. If breached, it can't be rolled back. You can't change how you walk. It's a permanent biometric identifier, just like a fingerprint or iris scan.
Contrarian: The Blind Spot of the Privacy Narrative
The typical reaction is to call for a ban on gait recognition. But that misses the deeper narrative. The real story is not the technology itself, but the economic incentive structure. Flock's business model relies on recurring subscription fees from police departments and homeowners associations. The more data they collect, the more valuable the platform becomes. The 69 prompts are a feature, not a bug. They are designed to increase lock-in by making the system indispensable.
Here's the contrarian angle: The emergence of such centralized surveillance systems could actually accelerate the adoption of decentralized identity and privacy-preserving technologies. Just as the Snowden revelations fueled the growth of encrypted messaging, the revelation of mass gait surveillance could drive demand for zero-knowledge proof-based identity systems. Imagine a world where you can prove you are a human without revealing your unique movement pattern. Where you can walk through a city and your gait is not recorded.
But there's a catch. The blockchain community has been obsessed with pseudonymity, not anonymity. On-chain analysis can already link wallets by transaction patterns. The same pattern-matching logic that identifies a gait can identify a wallet's spending habits. We are building the same surveillance infrastructure in crypto, but calling it "analytics."
I've seen this movie before. In 2022, during the bear market, I wrote about modular blockchains as a survival mechanism. The narrative was about resilience through fragmentation. The same principle applies here: The only way to resist surveillance is to fragment the data. Gait signatures should never be stored in a single silo. They should be distributed, encrypted, and only accessible with explicit consent.
Connecting the dots before they connect you.
Takeaway: The Next Narrative Shift
The 69 prompts are not just a technical artifact. They are a signal of a larger shift in the battle between centralized surveillance and decentralized privacy. The next narrative will not be about which layer-1 blockchain wins. It will be about who controls the data that defines your physical identity. The crypto community has a chance to build the infrastructure for self-sovereign identity that works offline as well as on-chain. But it requires a shift in focus from financial speculation to personal sovereignty.
The question is not whether gait recognition is good or bad. It's whether we are building systems that give individuals the ability to opt out. Because once your walk is recorded, you can't take it back. It's a forever asset, and it's not yours.