Alignment is a real thing, though it can be a less than apt description. Mostly the goals you write out and the goals you want to happen not being the same thing. You can also have alignment issues where the dataset doesn't match the real data in some way and the classifications break down. (I am not an AI defender I am very anti-AI)