The right question isn't whether AI makes mistakes. It's what happens when it does, and who is protected when it happens.