Berlin-based artist and researcher Simon Weckert has developed an experimental shirt designed to confuse some AI-powered surveillance cameras.
Called “Digital Camouflage,” the unusual garment looks like a bright Hawaiian shirt covered with colorful shapes and patterns. However, the design is not only meant to attract attention. Its main purpose is to test how well artificial intelligence can identify people in public places.
In demonstrations using the open-source YOLO object-detection system, the software detected nearby pedestrians and marked them as people. However, it did not always recognize the person wearing the Digital Camouflage shirt.
The wearer remained fully visible to the human eye. The computer system, however, sometimes failed to understand that it was looking at a human figure.
How Does the Digital Camouflage Shirt Work?
Modern security cameras can do more than record video. When connected to AI software, they can study live footage, identify people and objects, and sometimes examine activity in public places.
However, an AI system does not see a person in the same way a human does. It learns from large collections of training images and searches for common visual signs linked to the human body. These signs may include the shape of the head and shoulders, the size of the arms and legs, the position of the torso, and the outline of the full body.
Weckert designed the shirt to disrupt those visual signs. Its bright color changes attract the AI system’s attention, while its overlapping shapes make the wearer’s body outline harder for the software to follow. As a result, the system may fail to connect the head, arms, and torso as one human figure.
To a person, the design looks like a bright and unusual fabric pattern. To certain AI models, however, the same pattern may make the wearer difficult to classify. Interestingly, Weckert used AI while developing the shirt.
The design was created through an “adversarial loop.” In this process, a pattern was generated and then shown to an object-detection system. The system’s ability to detect the wearer was measured, and the pattern was changed based on the result.
This process was repeated until the AI’s confidence in detecting a person dropped. Weckert said this method was needed because humans cannot easily guess what an AI system will fail to see. Testing each design directly against the software helped reveal its weak points.
He then demonstrated the final pattern using YOLO, a widely used open-source system that can detect objects in images and videos in real time. Weckert created Digital Camouflage in response to plans for AI-supported video surveillance at Kottbusser Tor, a busy public area in Berlin.
The project is meant to start a wider discussion about the use of automated surveillance in public spaces. Weckert wants people to ask how these systems work, how accurate they are and who is responsible when they make mistakes.
As AI cameras become more common, they may be used to identify people, follow movement or study behavior. However, the public may not always know how the software reaches its decisions or how often it produces a wrong result.
Digital Camouflage shows that even highly advanced systems can have limits. Weckert has made it clear that the shirt does not make a person invisible and is not guaranteed to work against every AI camera. The tests involved the open-source YOLO detection system. He does not claim that the shirt can defeat any specific police, government, or private surveillance system.
Its performance may also change because of lighting, camera angles, distance, software updates, and the type of AI model being used. Digital Camouflage is therefore an art and research project, not a promise of complete anonymity.
Other designers have also tested clothing against surveillance cameras, including garments fitted with infrared lights. However, Digital Camouflage takes a different approach by targeting the way AI software reads shapes, colors, and body outlines.
The project offers a clear warning: AI surveillance technology may be powerful, but it is not always as reliable or perfect as it appears.
