Adversarial Fashion was a clothing line printed with fake license plates, intended to inject junk data into Automated License Plate Readers.
Automated license plate readers sit on police cruisers, tow trucks and street poles. They scan whatever is in front of them and try to find a plate in it. They are not especially good at this. Left alone, they already log fences, billboards and the lettering on the side of a van.
The clothes gave them something to find. Each garment carries dozens of plates that hold up well enough to a reader working at highway speed. The reader captures them, stamps each one with a time and a location, and drops them into the same database as actual cars.
Do they really work?
These shirts were developed and tested in the field on APIs like OpenALPR and mobile ALPR products, both static image and live video. After all these years, they still work! A surprising result in a time when computer vision models are continuously updated. I don’t (yet) have access to the models that Flock or Axon use to try them out. If you do, let me know.
Are shirts really going to make a difference against Flock or other ALPRs?
One shirt is not going to bring down a surveillance network, and I have never claimed otherwise. What the shirts show is that nobody is watching the front door of these systems. They take what they are handed, there is no meaningful check on the way in, and a dataset is a great deal cheaper to pollute than to clean. We see time and again that a system that will read a tshirt in as a car can and often does misread real plates, resulting in cases of mistaken identity that have already put people’s lives in danger.
In this way, anti-surveillance art projects like Adversarial Fashion are best at helping us tackle large unwieldy ideas like AI-enabled mass surveillance and making them digestible for the average person. They form concrete examples that let us better understand and talk about how these systems work, and figure out together the place they should or should not have in our lives.