How a Typo in a License Plate Turned a Driver into a Car Thief: The Flock Camera Failure

Flock's automatic license-plate reading cameras are deployed by police all across the United States. The story of one writer whom the system mistook for a car thief, tracked down and got arrested went viral, and it vividly illustrates what automated surveillance turns into at the slightest inaccuracy in the underlying data.

The Missing Digit

A dealership in Los Angeles reportedly had a car stolen that carried the New Jersey plate 34 03 DTM. But when the police report was filed, the plate was entered into Flock simply as 34 DTM — just the five large characters, without the small digit in the middle. And that is exactly how the algorithm “saw” it: it failed to read the nonstandard small digit and started firing off alerts to police the moment it spotted a Range Rover on the road.

The problem turned out to be far bigger than one car. A great many vehicles in Jaguar Land Rover's media fleet carry New Jersey manufacturer plates with the same structure — 34 ## DTM. Which means that every such car came under suspicion anywhere the local police had a contract with Flock. In that one week in Minnesota alone, four more cars with plates of the form 34 ## DTM were tracked; the writer just happened to be the first to get caught. There was only one way to stop it — to make the Los Angeles police correct the original report and update the data in the system.

“Everything Worked Correctly”

Flock responded to the wave of criticism by claiming that its systems had functioned correctly, shifting the blame onto the police. According to a company representative, the machine learning had honestly read what it had been told to: the characters 34 DTM were entered on the list, and that is what the system found. It was asked “is this here?” — and the answer was “yes.” It did not go on to check whether there was anything else nearby.

What's more, even if the unlucky digit had been of normal size, the match would still have triggered — the system was configured that way at law enforcement's request. Police often have to launch a search based on an incomplete plate, so they want an alert whenever any of the entered characters match. Verifying whether 34 10 DTM really is the sought-after 34 DTM is up to the officer.

Not Just Cars

Separately, the head of Flock apologized for having earlier called “terrorists” the activists who track the locations of the company's cameras — although observers took this as a change in PR strategy rather than a change of heart.

Meanwhile, according to data reviewed by the outlet 404 Media, police use the Flock network to search for people too, not just cars. Among the queries: “large man in a black-and-white hat,” “person on a skateboard,” “person in an orange vest and a hard hat.” Sometimes they reference the target's race or signs of their political views. Like any police surveillance technology, this one is not without its abuses.

Source: Schneier on Security