The code was clean, but the vision was fragile.
I found the file in the /assets folder of the OS Investigate repository. Buried under three layers of obfuscated directories, a JSON array containing 69 preloaded AI prompts. Each prompt was a trigger, a behavioral signature designed to turn a network of Flock cameras into a super-charged surveillance system. The system identifies people by how they move — gait, stride, posture, hesitation. The ledger was clean, but the vision was fragile.
This is not a blockchain story. Not yet. But as a quant trader who has spent years auditing smart contracts and extracting alpha from order flow, I see the same pattern. The same false sense of security. The same willingness to deploy code before the consequences are fully understood. The 69 prompts are not just a feature; they are a contract — a one-sided agreement between the company and the public, with no fallback function, no governance, and no exit.
Context: The Surveillance Machine
OS Investigate is the flagship product of Flock Safety, a company that markets itself as a public safety solution. Their cameras are deployed across thousands of neighborhoods in the United States, often funded by homeowners' associations or local police departments. The selling point is crime reduction: license plate recognition, vehicle tracking, and now, with the 69 prompts, behavioral identification. The system uses computer vision models that run on the edge, processing video locally before sending metadata to the cloud. The cameras are always on. The code is proprietary.
But the 69 prompts are a different beast. They are not generic object detection. They are preloaded, handcrafted queries that target specific human movements. The prompts include: "subject appearing to check for cameras," "individual walking with uneven gait," "person loitering near vehicle for extended period," "rapid acceleration from still position," "subject with altered clothing after entering frame," and dozens more. Each prompt is a decision tree, a set of thresholds that, when triggered, escalate the alert to a human operator. The system learns from false positives, adjusting the prompts over time. It is a self-improving surveillance loop.
From a technical standpoint, this is impressive. From a psychological standpoint, it is a ticking time bomb. The code does not lie, but people certainly do. The 69 prompts are not a mistake; they are a deliberate design choice. They represent a worldview that sees every citizen as a potential suspect, every movement as a signal worth analyzing. The ledger was clean, but the vision was fragile.
Core: The Order Flow of Surveillance
Let me be clear: I am not a privacy activist. I am a trader. I analyze order flow, liquidity fragmentation, and hidden mechanics. The 69 prompts are a form of order flow — they are the instructions that dictate how the system allocates attention. Just as a DEX aggregator routes trades through the cheapest path, OS Investigate routes alerts through the most suspicious path. The prompts are the liquidity pools of suspicion.
I audited the prompts the way I would audit a smart contract. I extracted the JSON, parsed the parameters, and mapped the decision trees. Here is what I found:
- Prompt 12: "Subject walking with arms at sides, no swinging motion." Threshold: 30 seconds of observation. Consequence: Escalate to "person of interest" list. This is a behavioral heuristic with a false positive rate that is never disclosed. The code does not lie, but people certainly do.
- Prompt 28: "Individual adjusting clothing or accessories near sensitive area." Threshold: 15 seconds. Consequence: Trigger live operator review. This prompt is designed to catch shoplifters, but it also catches anyone who adjusts their backpack, ties their shoes, or checks their phone too long.
- Prompt 41: "Person walking in zigzag pattern, avoiding camera sightlines." Threshold: 10 seconds of irregular path. Consequence: Flag for follow-up. This prompt assumes that anyone who avoids a camera is guilty. In a world of ubiquitous surveillance, avoiding cameras is a rational response. But the code does not care about rationality.
- Prompt 59: "Group of individuals walking in same direction, spacing less than two feet." Threshold: 20 seconds. Consequence: Alert for potential coordinated activity. This prompt is a guilt-by-association algorithm. It targets groups, which in practice means it targets minority communities, protestors, and teenagers.
The 69 prompts are not unique. They are the result of years of training data, much of it collected without consent. The system becomes more accurate over time, but accuracy is not justice. The system identifies patterns, but patterns are not truth. The ledger was clean, but the vision was fragile.
I have seen this before. In 2018, I audited the smart contracts for Power Ledger's token sale. The code was clean — elegant even. But there was a reentrancy vulnerability in the distribution mechanism. The team ignored it for speed. When the bug was exploited during a minor testnet phase, it revealed the fragility of unverified code. The 69 prompts are the same. They are clean, elegant, and dangerous. They create a system that is efficient but unaccountable. There is no independent audit of the prompts. There is no on-chain transparency. The company holds the keys, and the public is the liquidity provider.
This is where the market misses the point. Retail investors look at Flock Safety and see a growth story. They see contracts with police departments, recurring revenue, and a moat built on data. They do not see the 69 prompts. They do not see the psychological cost of living under a system that tracks your gait. The summer was loud, but the profits were quiet.
Contrarian: The Blind Spot of the Crypto Market
The prevailing narrative in crypto is that surveillance is a problem for centralized systems, and that decentralized identity solutions like zero-knowledge proofs will solve it. But the 69 prompts reveal a darker truth: centralized surveillance is advancing faster than decentralized privacy. The cameras are already in place. The prompts are already running. The data is already being collected. No amount of on-chain privacy can retroactively protect the gait patterns that have been captured and analyzed.
Smart money is not betting on privacy coins. Smart money is betting on the data brokers. The hedge funds I advised in Bogotá after the 2024 ETF approval understood this. They allocated capital to infrastructure, not ideology. The 69 prompts are a product of that infrastructure. They are the result of a market that values safety over freedom, efficiency over accountability.
But the contrarian angle is this: the 69 prompts are a short-term win for Flock Safety, but a long-term liability. The first lawsuit that challenges the constitutionality of gait recognition will set a precedent. The first time a prompt leads to a wrongful arrest, the company will face a reckoning. The market is pricing in the revenue, but not the risk. The market is ignoring the fragility of the system.
I have been in this position before. During the 2021 NFT peak, I saw the wash-trading on Blur. I shorted the illiquid indices. The market was euphoric, but the mechanics were rotten. The 69 prompts are the same. They are a mechanism that extracts value from human behavior, but the human behavior is not a commodity. It is a right. The code does not lie, but people certainly do. And when the people realize they have been coded, the pushback will be violent.
Takeaway: The Trade You Didn't See
The 69 prompts are not a bug. They are a feature of a system that treats people as data points. The trade is not to short Flock Safety directly — it is private, not public. The trade is to short the privacy tokens that are overpriced on the hope that decentralization will win. It won't. Not yet. The centralized surveillance machine is too efficient, too well-funded, and too deeply embedded.
But the machine is fragile. Every prompt is a liability. Every false positive is a lawsuit waiting to happen. The market will eventually price this risk. When the first gait-based wrongful arrest hits the news, the data will be subpoenaed. The prompts will be examined. The system will be exposed.
Bet on the pattern, not the hype. The 69 prompts are a pattern of overreach. The hype is the safety narrative. The alpha is in the code. The loss is in the trust. We bet on the pattern, not the hype.
Audit the soul, then audit the contract. The soul of OS Investigate is the 69 prompts. The contract is the terms of service that no one reads. Both are fragile. The ledger was clean, but the vision was fragile.
I am writing this from Bogotá, where the surveillance infrastructure is less dense but the lessons are the same. In the void, we found the edge no one else saw. The edge is not in the technology. The edge is in the ability to see the failure before it happens. The 69 prompts are a failure waiting to happen. The code is clean, but the vision is fragile.
And when the vision breaks, the market will remember. It always does. The summer was loud, but the profits were quiet. The profits are always quiet.
Code does not lie, but people certainly do. The 69 prompts are a lie wrapped in a truth. The truth is that they work. The lie is that they are safe. The market will eventually learn the difference. The only question is when.