The camo trick meant to fool AI drones may make trucks easier to spot

A paint job aimed at a target that can't be jammed

The Economist's *How to hide from killer drones* describes a low-tech answer to a high-tech problem: soldiers repainting vehicles in dazzle-style stripes and blotches, the same visual-disruption trick the Royal Navy used on WWI convoys, now aimed not at a human eye but at the object-detection model steering an autonomous attack drone. The distinction matters. One reader in the thread pointed out that the trick is old news against a human pilot flying a drone by video feed; what's new is a drone that decides who to hit without radioing a person first, a target that can't be jammed, spoofed by radio silence, or talked out of a strike, only fooled visually or not at all.

The piece's own thesis, reproduced word for word in the thread by a reader who quoted the article's closing line in full: "The probable result will be an arms race pitting increasingly sophisticated machine vision systems against cleverer and cleverer methods for fooling them." Camouflage, in other words, is entering a software fight that resets every time the underlying vision model gets retrained.

The fastest place to test that thesis in practice is the drone war being fought in Ukraine, where thermal and infrared imaging already do most of the targeting work once daylight fades, camouflage or none.

Even the dazzle-camouflage analogy needed correcting first

Before the thread argued about whether the paint works, it argued about what dazzle camouflage ever did. A commenter with over 3,000 karma responded to the WWI parallel with a correction:

"Dazzle isn't meant to hide. Dazzle is meant to confuse a human trying to estimate course and speed in order to send an unguided torpedo at it. You need to look at hit rates to judge effectiveness."

That reframes the entire modern experiment: the WWI trick worked, if it worked, against a person doing mental trigonometry, not against a machine measuring pixels.

The stripes make the model's job easier, not harder

A commenter with over 14,000 karma dismissed the article's framing outright:

"This is an odd article that tries to elevate some random grunt in the field painting their truck white stripes to grand battlefield strategy in the face of autonomous AI killer drones... it obviously is not effective, not least because the drones it's talking about barely have the resolution at altitude to resolve that detail."

A commenter who linked out to Ars Technica's and the BBC's reporting on Ukraine's drone tests pushed back that field evidence was starting to cut the other way, with autonomous targeting already working well enough to matter in combat.

The sharper technical objection turned the paint into a liability rather than a defense. An engineer who proposed a close-in weapons system instead argued:

"Even civilian LLMs recognize that the object on the photograph is a military truck, except they can't explain why it's been painted to resemble a zebra. Most dedicated machine vision models easily lock in on a boxy shape moving along a road. If anything, the stripes make the trucks easier to see."

His proposed fix was CIWS, a close-in weapons system built to shoot down multiple incoming threats from any direction, which he called the "real answer to killer drones."

A CIWS pitch that loses to a cheap drone swarm

The CIWS idea didn't survive contact with the thread's cost math. Another engineer in the thread laid out the asymmetry:

"A CIWS can only fire at one direction at a time, so 2+1 drones has an extremely high chance of taking it out for around $2500 or less. CIWS is multiple magnitudes more expensive."

An electrical engineer who works in UAV and robotics narrowed the objection further: fixed-wing drones fly in predictable lines a rotating gun can track, but multirotor drones maneuver, and a swarm of them overwhelms a system that can only aim in one direction at a time.

Heat and radar were never fooled by paint anyway

A commenter with close to 8,500 karma placed the whole debate inside a much older one: infrared targeting has nothing to do with what a vehicle looks like.

"Saddam used USSR anti-ship missiles, old even then, to attack Iran oil storage tanks at night as the missiles were easily able to lock on that large bright IR emission of the tanks still hot from the day against the cold night desert."

The lesson predates autonomous drones by four decades: paint changes shape, not heat. An engineer who linked to a radar module built for drones added a third sense paint can't touch at all: strap it on, and the drone reportedly sees through walls, no visible light required.

Put the three objections together and the article's premise gets narrower than its headline: dazzle paint might buy a fraction of a second against one specific, visible-light, altitude-limited camera, while the sensors most likely to do the targeting at night were never looking at color to begin with.