AI cameras are trained to recognize people, so an artist designed a T-shirt to fool them
Simon Weckert created a pattern that fails to record the wearer as a 'person'
What if surveillance cameras automatically decided details of an environment without recording the actual events going on? A German artist has designed what is equivalent of Harry Potter’s invisibility cloak to deceive AI models into not detecting a human even though one is present. Berlin-based artist and technologist Simon Weckert calls his project “Digital Camouflage.” As part of this project, he has designed t-shirts that will make a human invisible to AI surveillance cameras. The story was reported on September 1, 2026, by 404 Media, and has gained 851,000 views and over 64,000 likes since.
Countering surveillance in style
Weckert developed these colorful, patterned Hawaiian-style t-shirts in response to the increasing implementation of AI-based public video surveillance in Berlin. The t-shirts feature an abstract design pattered in bright pink, orange, and green blobs. According to him, this pattern specifically disrupts the visual continuity and statistical patterns of computer vision models to recognize human shapes. Weckert tested the garment against YOLO (You Only Look Once), an open-source object detection system. While the person itself doesn’t disappear to the AI, the model fails to register it as a human.
In a demonstration shared by the outlet in an Instagram reel, Weckert showed how the AI camera detects a person using image-recognition software. Each time the system registers a person, it places a green box around them and labels them as “person.” When Weckert placed the new shirt in front of the person’s body, the box disappeared, and the system was disrupted and hence failed to register it as a person.
Interfering with computer vision
The hint probably lies in the visual pattern of bright colors and overlapping shapes that confuse the AI system into refusing to register the wearer as a person. Weckert revealed that he created this pattern with the help of AI, in something he calls an “adversarial fashion,” a fashion design to confuse and trick surveillance cameras.
Weckert, however, is not the first designer of a clothing like this. Researcher Kaidi Xu and colleagues at Northeastern, MIT-IBM Watson AI Lab, and MIT previously designed a t-shirt that tricks a variety of object detection models into failing to spot people, per Northeastern Global News. While testing the pattern, two men walked side by side towards a camera, one dressed in all black and other in khaki and a white t-shirt with a brightly-colored abstract pattern centered around on the front. The AI analyzing the camera recording detected only one person. Xu and his team found that the wearer was able to evade AI detection more than 60% at a time.
Awesome yet tragic
While viewers appreciated the idea, they also showed remorse about this being the last resort. @mlot11 speculated, “Great. Now they will just train the model on that shirt, and it won’t work anymore on the next version release.” @reis369247 said, “Now, they will train AI specifically for this pattern. Next time, don't advertise it.” @chappellruin joked, “Can I get it in a jumpsuit? So cool!
More on Scoop Upworthy
Woman fell in love with an AI chatbot. She just updated her relationship status
Share on Facebook
