Sounds like we're talking about different kind of safeguards. I mean stuff like a character in a game ending up in a broken case due to something that is theoretically possible but very unlikely or where the consequence is not worth the effort. An AI has no problem taking those into account and write tedious safeguards, while I skip it.