I imagine that if the structure of a human brain was well-understood and mathematically describable the way NNs are, generating adversarial inputs would be completely feasible.