Product1 distinct publisher3 min readPublished
Simon Weckert tuned a garish pattern against an open-source detector until it stopped seeing a person at all, which puts the weak point in the one stage of a vision pipeline most teams inherit and never test.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
An operations screen during a behaviour-scanning pilot shows detections, tracks, alerts, whatever the dashboard has been told counts as an incident. Misses are not a row in that table. Because the recognition stages run on whatever the detector hands them, a person the detector never boxes produces no downstream output at all [13], so the failure Weckert is demonstrating is silent by construction.
The mechanism is unglamorous. Detectors key on edges, contrasts, limb proportions and silhouettes learned from millions of training images [5], and the account in Dexerto has the shirt doing two things to that: saturated colour overloading the early perception layers, and overlapping shapes breaking the outline so the model never binds arms, torso and head into one figure [6]. Weckert's phrasing beats most vendor copy on this point, describing a pattern that "shouts over the quiet statistical whisper that says 'person'" [7].
Worth being clear about provenance. This all comes from Weckert by way of Dexerto, which credits Futurism for the explanation of where in the pipeline the attack lands [4]. The tuning was done against YOLO, a single open-source detector, in a generative loop run until the model's confidence in seeing a person collapsed [8]. He says he checked the result against a live camera at Kottbusser Tor in June 2026 [10], roughly two months ahead of the August 2026 pilot rollout he built it to answer [14]. There is no independent measurement of the deployed system's recall in the material.
What gets pitched is behaviour reading. What gets built is a chain, and stage one of that chain is usually a stock detector on default weights, picked because it performs well enough on people who dress like the training images. Plenty of people do not. They wear coats that swallow the silhouette, and now some of them can buy an $88 shirt [1] that Weckert says was designed to hold up through fabric folds, shifting light and changing angles rather than only in controlled tests [9].
The sorting question for your own pipeline has two axes: which stages did you tune versus inherit, and which stages, when they fail, write a line somebody can review versus nothing at all. Inherited and silent describes stage one for almost every team shipping vision today, and that quadrant is where the risk sits unpriced. The recommendation carries a bill: treat detection as a stage you own and measure adversarially, and accept that a detector loosened enough to catch outline-breaking patterns will also bind more non-person shapes into person boxes, since it is keying on the same cues [15], and a human pays for each of those in review time. The thing to ask a vendor is which stages have ever been measured against someone actively trying to break them. Weckert's own framing is that the garment is "not a tool against the police, it's a test image for all of us" [11].
Ranked by verification strength, evidence, and original report placement.
German artist Simon Weckert designed a garish Hawaiian-style shirt called Digital Camouflage, costing roughly $88 and covered in saturated pink, orange and green shapes, which he says makes the wearer invisible to AI-powered surveillance cameras.
Weckert built the shirt in direct response to Berlin's August 2026 pilot rollout of automated, behaviour-scanning AI surveillance at Kottbusser Tor.
Weckert wrote on Instagram: "The system reads the behaviour of everyone passing through, so I made a shirt it cannot read."
According to Futurism, the shirt targets the first stage of AI detection, the point where software tries to draw a box around a human shape before running facial or gait recognition.
AI models do not understand what a person is; they rely on statistical patterns, edges, contrasts, limb proportions and silhouettes learned from millions of training images.
The shirt's jarring colours overload the early perception layers while its overlapping shapes break up the body outline, so the software never binds the arms, torso and head into one recognisable figure.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 30, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
From $50 to $2.25M: Niu Lai prices human authorship, not AI reputational risk2 distinct publishers
product
Cinemas, classrooms and ICE: smart glasses now need a venue-policy contingency1 distinct publisher
product
Roblox counts 28 clicks, 21 of them purchases, as an eight-year-old's waiver1 distinct publisher
product
HaveIBeenFlocked turns Flock's own audit logs into a plate lookup anyone can run1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
One interested witness, no reproduction
Trace any technical sentence here and it ends at Weckert: the adversarial loop, the collapsed confidence score, the June test in front of a working camera. Dexerto credits Futurism for the pipeline explanation, so even the mechanism is relayed rather than checked. The single detector named, YOLO, comes without a version, a confidence threshold, or an image anyone else can score, and Berlin's actual pilot stack is never described at all. The explanation of why the trick should work is sound; the claim that it did work is unexamined.
The surveillance has a rollout date; the countermeasure has a price
The only thing here with real deployment behind it is the thing being evaded: behaviour-scanning cameras said to be live at Kottbusser Tor since August 2026. On the other side of the lens sits one garment at $88, worn once for a test by the person who made it. No second wearer, no sales figure, no transit authority conceding a missed detection, no researcher rebuilding the pattern. As uptake, this is a demonstration, not a diffusion.
Invisibility promised, one detector actually beaten
'Invisible to AI surveillance' is a much larger claim than the work supports. What was demonstrated is that a pattern tuned in a loop against an open-source detector stopped that detector scoring a person, plus one unverified pass in front of a Berlin camera two months before the pilot began. Weckert's own framing is the more defensible version — a test image, a question about how much trust a system deserves — and it survives the gap better than the headline does. Nothing here is fabricated; the reach exceeds the measurement.
Everyone in the chain wants this to be true
The person supplying every technical detail also sells the shirt, and his record — 99 phones in a handcart to fake traffic on Google Maps — is a career built on legible provocations. That is not disqualifying, but it sets the direction of any error. Downstream, Dexerto's entertainment desk is aggregating Futurism on a story whose appeal is the claim itself; a null result would not have travelled. No one in the chain is positioned to run the test that could fail, and no one asked Berlin police or the pilot's vendor for the version that might.
Sure about the sourcing, unsure about the shirt
We can be firm on the shape of this: one publisher, one voice, one detector, a test that predates the deployment. We can also be firm on the structural point, which does not depend on Weckert being right — perception-layer fragility and the false-positive cost of retuning follow from how detection works. What we cannot judge is whether the shirt defeats Berlin's live system, and with no contradicting account, silence is not corroboration. Confidence would move quickly on a single independent trial against a named model.