Cost the company nothing

The mistake

The bot-check challenge shows users a grid of photos and asks them to pick out a specific object, relying on humans still beating automated vision models at recognizing real-world images. Here the challenge photos themselves are generated by a model instead of sourced as real photographs, so they come out as indistinct, low-fidelity blurs that don't clearly depict the target object at all. With no genuine visual signal to key off, human users can't solve it reliably, while the bots pass through unaffected, which defeats the entire point of the check.