You could say that all perceptual illusions are adversarial inputs for the human senses. They're just not damaging ones.
In a lot of adversarial input stuff for AI, you're trying to get the AI to misclassify something, not necessarily to damage the AI or even change its learning at all. In Snow Crash the perceptions were definitely damaging to the perceivers, but I don't think you have to go nearly that far to have adversarial inputs!