The garment looks like a box of paints exploded onto a soccer jersey, bright reds, blues, and yellows popping out from the cotton fabric. Wearing it in public would surely attract some looks, but not from AI. That's because the shirt I am wearing is the latest prototype from Cap_able design, an Italian fashion studio that produces clothing engineered to thwart surveillance technology.

Video surveillance enhanced by artificial intelligence has exploded in recent years, projected to grow from a $3.90bn industry in 2024 to $12.46bn by 2030. As the capabilities and reach of such technology continue to expand globally, individual designers and startups like Cap_able dedicated to helping people find at least one more way to resist: fashion.

Cap_able founder and CEO Rachele Didero started developing anti-surveillance fashion technology in 2019 while studying at the Fashion Institute of Technology in New York. She said she had become increasingly aware of biometric cameras proliferating around the city, and enlisted a friend working in computer science to do something about it.

“The problem with [surveillance] technology is that citizens don't have the possibility in the physical world to opt out,” she said. “In the digital space we can say yes or no to cookies. In the physical world, we don't have this option. We wanted to create this physical protection that would allow the citizens to give their consent, or not.”

Cap_able garments are constructed to be reversible, so the user can choose when they get dressed in the morning if they want to be detected by AI, or not.

AI vision systems are trained on massive datasets to recognize categories of images. Adversarial images – pixels that alter the computer's ability to recognize such images – have been developed over the years by various academic researchers and corporations. The new process patented by Cap_able algorithmically develops adversarial patterns, creating “AI-camouflage” that misleads computer vision systems. Cap_able then translates those digital adversarial features into knit structures through a textile engineering system that maintains constraints including gauge, yarn behavior, jacquard structure, and color limitations.

The knits are less stretchy than your average vintage sweater–they have been engineered to produce less give in the fabric to avoid altering the technology. When these adversarial patterns are translated into fabric, they are able to lower the AI’s classification confidence or push it into the wrong categories. It’s like fogging up the computer’s glasses when it’s trying to focus on an object.

Today, the company sells nine different garments including sweaters, shirts, dresses, and pants with her patented designs, with prices ranging from 127€ to 670€ and more products in development. Cap_able creates the final product with a specialized knitting company in Italy using high-quality materials. The quality of the products is clear–they look and feel luxurious. And what’s more, they’re cool. Wearing Cap_able’s multi-colored sweater — meant to trick computers into identifying a dog — attracted comments from strangers, including one person who complimented my “dog sweater,” suggesting the prints can trick humans, too.

“With artificial intelligence increasingly being linked to biometrics, we don’t have many regulations to protect us,” Didero said. “We live in a world where every step we take is recognized and collected, and it is very important we create more awareness and protection.”

AI-assisted surveillance is increasingly popular everywhere. In Paris, where I live, hundreds of AI-enabled surveillance cameras scan crowds to detect abandoned packages, weapons, and other potential terrorism risks, after being implemented for the 2024 Olympics in a move the Oxford Human Rights Hub said risks “eroding privacy and civil liberties.”

The city of London has used mobile facial recognition cameras attached to cars to search for individuals involved in crime, scanning approximately 4.7 million faces over the course of the year. In the United States, police forces are increasingly using AI-equipped cameras for real-time alerts and suspect detection. The US Government Accountability Office published a report identifying seven agencies using such methods in 2024.

In the face of the increasing ubiquity of surveillance technology, companies like Cap_able are finding more and more customers. Didero said she has seen an increase in sales in recent years, which she attributes to the AI boom and an accompanying rise in public consciousness surrounding its harms.

Cap_able is not the only provider of such adversarial fashion. Nothing Personal decided to test other clothing providers including AntiAI Clothing, Yelo Pomelo, Cap_able Design, Oogaly, and RiotAssembly. I pulled garments from their respective boxes and marveled at the variety of colors and patterns that can potentially trick AI. The AntiAI clothing included multiple t-shirts, including a black and white pixelated pattern, a pattern that resembled a blurred, colorful photo, and a multi-colored psychedelic pattern. Some were loud, and honestly not to my sartorial taste–while others were surprisingly wearable and could easily become wardrobe staples.

Skeptics of anti-surveillance fashion have noted that widely-varying and rapidly-evolving algorithms can quickly render previously functional adversarial patterns less effective. For example, Berlin-based technologist Adam Harvey developed multiple projects that are less functional years later. As he noted on his website: “Camouflage, in general, should be considered temporary, but especially technical camouflage that targets quickly evolving algorithms.”

Didero said Cap_able tests their products with the widely-deployed object detection algorithm called YOLO (You Only Look Once), which is widely considered to be the fastest and most accurate identifier. Tyler Brownell, founder of AntiAI Fashion, tests his products for efficacy with Google’s TensorFlow API, an open-source neural network trained for different kinds of object detection, including of humans.

Brownell, who has a degree in computer science, said he fell down a rabbit hole of privacy rights around 2022. Soon after, he started creating products integrating modified adversarial patterns he found online. Many anti-surveillance garments were prohibitively expensive, so he created his company and started developing his own patterns. Today, AntiAI Fashion offers 44 products on its website, topping out at $83 for a hoodie, and has shipped more than 7,000 orders. He has seen an increase in customer interest in recent years, particularly in the European market.

Brownell acknowledges that since founding his business in this space three years ago, the algorithms are getting more difficult to trick. The aim, he said, is to create an “empowering piece of clothing crafted with specific goals in mind” while noting that he is “not saying it will always work.”

Social media firms and data brokers collect vast amounts of online data about what we do, where we go, and who we know. Companies like Palantir integrate this data and more into what the ACLU calls a “centralized dossier on everyone,” permanently storing comprehensive digital profiles that track people through actions and locations, rather than just biometric identification that can be obscured with shirt patterns.

“There are certain costs to progression as a society–there was never really a social contract we signed where we agreed to mass surveillance and tracking,” said Brownell. “People deserve privacy and anonymity.”

Critics of adversarial fashion say because the surveillance state is so complex and wide-reaching, anti-surveillance fashion is more effective at raising awareness than actually changing the status quo. In his paper on the topic, researcher Torin Monahan suggests such garments represent an “aestheticization of resistance.” He added that marginalized groups experience surveillance differently and thus will be impacted differently by personal resistance efforts.

“It's not universal that we're all simply exposed to surveillance–some people's exposure is much more consequential than others,” he said.

Such targeting can have a more pronounced impact on more heavily-surveilled minority groups.

Another development complicating the deployment of such technologies is the rise of anti-surveillance measures now being appropriated by the state, as in the increasingly common tactic of ICE agents obscuring their faces while detaining people. In other words, anti-surveillance fashion and its impact is necessarily complicated by the identities of those who wear it, and the evolving technologies that shape its development.

“Practically speaking, I don't think that these technologies are going to be very effective at changing these power directions in society,” Monahan said. “But as consciousness-raising or awareness-building exercises, I think that they do have a purpose.”

Developers, customers, and advocates for anti-surveillance fashion cite awareness as a central component of their passion for the products. Andrew McGregor, an artist and researcher based in Los Angeles, joined me in October by video call, proudly wearing a Cap_able-designed sweater he got about a year ago.

“Privacy is a human right, and we have just given it away,” he said. “So many of us feel like we can’t do anything about it, and to me this kind of technology is a way of saying that even if the bastards are winning, I still care.”

McGregor said he would not rely on a sweater alone to avoid detection in “risky situations,” but on a day-to-day basis he finds it keeps privacy top of mind.

“When you are wearing one of these sweaters, you have a sense that you are somewhat camouflaged, and you start to notice more cameras all around you,” he said. “It makes me aware of how often I am being surveilled.”

He added that the conversations his Cap_able garments spark are perhaps even more important to him than their practical functionality. He wears his sweater several days a week and often turns compliments on the clothing into discussions with people about surveillance. In other words, he is ultimately wearing the sweater not to avoid detection from AI, but to be seen by people he might help educate.

“There is a symbolism to it–it’s like a public health measure,” he said. “When I can educate people while I am wearing these clothes, it feels like I am actually doing something to reduce the harm. For me, it's a reclamation of humanity.”