Comment by chroem-

Reinforcement learning and adversarial training can render both of those concerns as non-issues in practice.