At first glance, Simon Weckert‘s new Digital Camouflage collection looks like loud streetwear: a dense, abstract collision of colors, shapes and visual static. But the pattern is not there simply to stand out. It is designed to make computer vision systems less certain that a person is standing in front of them at all.

That claim needs a careful translation. The clothes do not make anyone physically invisible, and they are not a magic cloak that defeats every security camera. Instead, they are an experiment in adversarial fashion – garments engineered to exploit the blind spots of AI systems that identify people and objects through cameras.

Weckert’s idea is disarmingly simple: if AI surveillance increasingly reads the world as data, why not give it data it cannot easily read? His garments use a continuous, high-frequency pattern called an Adversarial Texture, or AdvTexture, across the full surface of the fabric. The goal is to feed a person-detection model confusing visual signals from multiple angles, weakening its ability to form a stable outline of the wearer.

In the language of machine learning, this is an adversarial attack. A tiny, deliberate change to an image can cause a neural network to misinterpret what it sees, even when the same image looks ordinary to a human. Put that idea on a T-shirt or jacket and the effect becomes oddly theatrical: you are still plainly visible to people nearby, yet the automated system looking at you may see visual noise, false features or something that does not quite register as human.

That is a meaningful distinction. Most public concern around AI surveillance focuses on facial recognition, which tries to identify a particular person. Digital Camouflage appears aimed earlier in the pipeline: person detection. Before a system can decide who somebody is, it must first decide that somebody is there. If it cannot confidently draw a box around a body in a video feed, subsequent analysis becomes harder.

The technical challenge is that clothing is a terrible canvas for precision computing. Fabric moves. It wrinkles, stretches, folds over itself and looks radically different under daylight, fluorescent lighting, rain, motion blur and a camera mounted high above a street. Older adversarial-clothing concepts often used a fixed patch, but a patch can vanish behind a fold or fall outside the camera frame. Weckert’s answer is to cover the whole garment in a seamless, tileable pattern rather than trusting one small printed target.

The collection uses a generative method described as TC-EGA to build that repeating textile design. It is then digitally printed on a fabric blend of 65% recycled polyester and 35% polyester, manufactured in Latvia. The result is less sci-fi invisibility suit than graphic, wearable computational interference.

There is a broader cultural point here, too. Cameras have always watched public spaces, but AI has changed what watching can mean. Modern systems can detect bodies, estimate attributes, track movement across frames and flag unusual activity at a scale no human security team could match. Anti-surveillance fashion turns an everyday personal choice – getting dressed – into a small confrontation with that automated gaze.

Weckert is far from the first designer to explore the territory. Artist and researcher Adam Harvey’s Computer Vision Dazzle used makeup and hairstyling to interrupt the facial patterns algorithms rely on. Other designers have tried retroreflective eyewear, distorted facial motifs, machine-generated knits and infrared-based approaches. The common idea is not to disappear from society, but to make automated recognition less frictionless.

Still, this is where the hype needs to stop. The success of adversarial clothing depends on the specific model, camera angle, lighting conditions, image compression and the way a surveillance system processes footage. An outfit that confuses one detector in a controlled demo may do very little against another system trained on different data. Systems can also be updated, reprocessed after recording, or supplemented with other signals such as faces, gait, phone data or additional cameras.

That does not make the project pointless. A good deal of design work is valuable because it exposes an assumption people had stopped noticing. In this case, the assumption is that machine vision is neutral, inevitable and always more capable than the people it watches. Digital Camouflage reminds us that these systems are built, trained and fallible – and that their weaknesses can become raw material for art, fashion and protest.

The clothes may never make their wearer truly invisible. But they make a more interesting proposition: perhaps invisibility in the AI era is not a vanishing act. Perhaps it is the ability to occasionally become difficult for a machine to understand.