Flock's cameras no longer need your license plate to follow you
A police camera network that made its name reading number plates now claims it can pick a person out by how they walk, what they wear and how they move. The code behind it ships with 69 ready-made AI prompts.
For years the pitch for Flock Safety was simple and narrow: cameras on poles that read license plates, flag stolen cars, help find a missing kid. A plate is a thing you can point at. It belongs to a vehicle, it can be checked, and if the police looked you up, there was a record with a number on it.
Decrypt reports that the code behind Flock's newer tool, OS Investigate, arrives with 69 preloaded AI prompts — pre-written instructions that let an officer search camera footage in plain language. Not just "find this plate," but descriptions: clothing, body type, the way someone moves. The company's own material frames it as identifying people without a name and without a plate. That is a different product wearing the same housing on the same pole.
A license plate is a key. You need the key to open the file. Take the key away and replace it with "a person in a grey hoodie who walks like this," and the door has no lock on it at all. You do not have to be registered anywhere, own a car, or carry ID to become searchable. You just have to have walked past a camera.
Surveillance used to cost labour. Following one person meant a car, two officers and a shift. That cost was the real limit on how often it happened — not the law, the payroll. An AI prompt costs nothing to run and can be run against everyone at once.
When watching gets cheap, the question stops being "who deserves to be followed" and becomes "why not everyone."
There is a second thing here, and it is quieter. Plate reads produce a clean record: this plate, this time, this camera. A description-based search produces a guess with a confidence score. Guesses are harder to audit and much harder to appeal. If a system decides your gait resembles the gait in a clip, you have to disprove a pattern, not a fact. We do not know how often these prompts are wrong, because that number has not been published.
Anyone who moves through a city on foot, on a bike or behind a wheel — which is most people. It lands hardest on the ones who are outside all day: couriers, delivery riders, rideshare drivers, night-shift workers, people whose job is literally being on the street for ten hours. Then there are students and twenty-somethings who go to protests, or to a clinic, or to a lawyer's office, and assumed the trip was private because nobody knew their name. Under a plate-based system, it mostly was. Under this one, a name was never the point.
Watch for three things. First, whether Flock publishes the full list of those 69 prompts, or whether it stays in the code and gets read by journalists instead. Second, whether any city council votes to pause or renegotiate its Flock contract once residents learn the cameras are no longer just reading plates — that is a public vote, on a public agenda, and it is the clearest signal of which way this goes. Third, the first court case where a defence lawyer asks how a gait match was scored, and what the error rate is. Whoever answers that question first sets the standard for everyone else.
The 69 prompts were not announced. They were found in the code. That tells you the capability shipped before the conversation about it did — and it raises a plainer question: how many of the cameras you already walk past every day are running software you have never been told about?
Sources: Decrypt, "Flock's New AI Police Tool Can Track Drivers Without a Name or License Plate" (decrypt.co), citing the underlying code and product materials for Flock Safety's OS Investigate.