A button-down shirt covered in blurry pink and green globs can make a person invisible to the same kind of object-recognition AI that Berlin police are now pointing at one of the city’s busiest transit plazas. The garment, called “Digital Camouflage” by Berlin artist Simon Weckert, drops the “PERSON” bounding box that person-detection models normally draw around a human figure. It exists, Weckert says, because the cameras it defeats are no longer hypothetical.
What the shirt actually does
Weckert’s project page documents the garment as “textile engineered to confuse the object-recognition algorithms used in surveillance” (simonweckert.com). When worn on the body, the shirt targets exactly the features these systems look for when searching for a human figure; in the project’s words, “the algorithm no longer registers a person.” In a demo filmed for 404 Media, a generic object-detection system placed a green “PERSON” box around the journalist; when Weckert held the shirt up in front of him, the box vanished. When he lowered it again, the box reappeared. The effect is repeatable and reversible.
Weckert’s documentation is careful about scope. The project page makes no claim about defeating every system under every condition, and notes that “Demonstrations use a generic open-source object-detection system (YOLO).” He links his project to three academic papers on adversarial clothing for person-detection evasion: arXiv 2510.17322, arXiv 2511.16020, and Hu et al.’s arXiv 2203.03373 on adversarial texture for fooling person detectors in the physical world. The research community has been publishing on this for years; Weckert has translated the lab results into something a person can buy and wear. The project’s documentation is published under a CC-BY-NC-SA 4.0 license, and per the site, “a share of the proceeds from every shirt goes to organizations defending digital civil rights and fighting the expansion of surveillance.”
Weckert framed the project as a direct response to a specific deployment. “Obviously people don’t like it because it means that AI is tracking the movements and behaviors of people,” he told 404 Media. He added that the cameras he wanted to push back against can detect when someone is lying on the ground, which he noted means “homeless people could be detected and police get triggered” (404 Media).
The Berlin cameras it was built against
The deployment that triggered the project is the new police camera trial at Kottbusser Tor, a long-contested transit plaza in north-central Berlin. Berlin Inspector reported in early August 2026 that the pilot was set to begin later that month, with Kottbusser Tor as the first site and Warschauer Strasse next on the list (berlinspectator.com). Berlin’s Interior Senator Iris Spranger, a Social Democratic politician, described the system to the publication as non-biometric: officers see “stick-figure outlines, not faces,” and what gets flagged is movement patterns rather than identity. Her published line, as quoted by Berlin Inspector, was: “The face will not be recognizable. What matters is the movement.”
The legal basis is Berlin’s General Security and Public Order Act (ASOG), amended in late 2025 (the berlinspectator.com piece pins the amendment to December of that year). The plan was discussed at the House of Representatives’ interior committee in March 2026. Cost projections from the Berlin Inspector report put investment at about 2.47 million euro and operating costs at about 2.13 million euro for the crime-hotspot cameras between 2026 and 2030. An earlier proposal to extend the same AI system to the parliament building itself was dropped after objections from lawmakers, which is itself a small but concrete data point on where the political line sits today.
This police trial is separate from a parallel expansion by BVG, Berlin’s public-transit authority. According to The Berliner, BVG already has 80 AI-aided cameras installed across 16 U-Bahn stations, with 24 at Schloßstraße alone, and is negotiating with Berlin’s data protection commissioner to extend coverage into public spaces above the platform level (the-berliner.com). In just over two months those systems generated 44 alarms about people in track areas and 6 about unauthorized access to operating facilities. The contract for the BVG work went to Adesso, a German IT firm The Berliner notes had a major cyberattack between 2022 and 2023 in which attackers gained substantial access to internal systems and data.
What This Means
Counter-surveillance fashion is not new. Adversarial-pattern glasses and T-shirts that defeat face recognition have circulated since at least 2010. What is different here is the specific response to a specific, named deployment. Weckert is not making a generic “AI is watching” statement. He is pointing at a camera installation that a city government has authorized, named the law that authorizes it, and shipped a product in the same month it goes live. The CC-BY-NC-SA license means anyone can study, modify, and reissue the pattern. Proceeds are routed to civil-rights organizations, which makes the project also a fundraising tool.
The limitations are real and worth being honest about. Weckert’s own page says “no pattern can promise to defeat every system under every condition.” The demo used one detector in one lighting setup. Berlin’s police system uses behavior-recognition software, not just person detection, and the vendor has not been publicly named in the Berlin Inspector report. Spranger’s framing, that the system does not identify faces, is technically consistent with what a behavior-recognition pipeline actually does, and is also exactly the kind of claim independent auditors will need to test against the live footage.
For readers outside Berlin, the through-line is portable. Object-recognition surveillance cameras are being installed in public spaces in cities around the world, usually with the same two-part justification: it does not record identity, and officers review anything flagged before any action is taken. Both claims can be true in narrow technical terms and still add up to a society where every public plaza is observed by software, 24 hours a day, on the assumption that anomalous movement is something the state should know about. A printed shirt is one of the milder counter-moves a person can make, and it is now commercially available.
The Bottom Line
Berlin’s first police-run behavior-recognition camera trial went live in August 2026, and a local artist responded with a CC-licensed shirt that makes person-detection AI drop the “PERSON” label. The deployment is real, the countermeasure is real, and neither side has solved the underlying question of whether software should be watching public plazas in the first place.