Comment by cout

How do we (or can we) know adversarial attacks don't work on us?