Well, it generally does take geniuses to achieve something monumentally stupid. Same principle as (unintentionally) creating an Unfriendly AI: current researchers are not competent enough to pose any reasonable risk of it, but a future genius might just fail hard enough.