The “noRecognition” project develops patterns to confuse AI recognition systems

With the rapid spread of AI-powered surveillance cameras, automatic recognition of people and vehicles has become a part of daily life in many cities and public places.

Facing this reality, American security researcher Bill Swearingen proposed a new project that aims to reduce the ability of these systems to track individuals through visual patterns specifically designed to confuse recognition algorithms.

31 million tests to find the right pattern

According to the German newspaper Focus, Swearingen launched his project under the name “noRecognition” based on his conviction that privacy is a fundamental right that must be protected.

To achieve this, it conducted about 31 million tests over an entire year to develop patterns that are difficult for artificial intelligence systems to correctly interpret.

The German Shape website explained that the development process relied on testing a huge number of designs and constantly modifying them, until arriving at patterns that achieved the highest percentage of confusion among the different recognition algorithms.

How do anti-surveillance clothing work?

These clothes do not prevent cameras from filming, nor do they make the wearer invisible to the human eye. But the patterns printed on them target the software responsible for analyzing the images.

According to the German newspaper Focus, these patterns may cause recognition systems to fail to classify the person or vehicle in front of them, which reduces the ability of automated systems to detect the target or issue alerts related to it.

Swearingen confirms that his designs succeeded in disabling eleven known automatic recognition systems, including Flock systems for reading vehicle license plates, Axon cameras installed on the bodies of police officers, in addition to the Clearview AI facial recognition system.

From Def Con to the markets

The Def Con conference in Las Vegas witnessed the first large practical demonstration of the technology, where a car was wrapped in one of the developed modes and driven in front of a monitoring system based on artificial intelligence. According to the developer, the system failed to recognize the car even though the camera continued to record it.

The German Shape website indicated that the experiment also revealed some technical challenges, as the wheels remained one of the elements that camouflage did not succeed in completely hiding it from the analysis systems.

A promising tool, albeit with limitations

Swearingen has already begun selling the designs on T-shirts and hoodies through a crowdfunding campaign, and in the future he plans to develop custom wraps for vehicles.

However, the effectiveness of this technology is not always guaranteed, as its success is linked to specific systems that have been previously tested.

As AI models are updated or different algorithms are used, their ability to mislead may decline. However, the project highlights increasing attempts to find a balance between technical progress and individuals' right to privacy.

Edited by: Efaq Benkirane