I know this comment is old but, and the model even changed, but... Saying "no X" usually works.

Try putting it in other ways in the prompt. Also, edit the character itself, the maximum amount of things you can change, even the ones at the "show more settings" part. Include different ways of saying what you want and what you don't want. There are, for example, the quite direct way ("do not include: 'list of items/things you don't want'"), and the more lengthy one ("I don't like X thing, so you must never even mention this word", and maybe complete with a ", use Y or Z instead").

This way, your problems will often be solved. Yet, there will be times in which you will have to generate again, among other problems. I have a problem, for example, in which in smutty roleplays the AI often has the problem of trying to shove foot fetish in it. I don't like it, so one day I wrote "no foot fetish (not even boot licking)", and the AI wrote "and you licked her boots (without fetishizing it)". Indeed, the AI wrote "without fetishizing it" between the round brackets/parentheses as if it solved everything.

That said, I never had the problem with the words you are talking about, so, maybe it's your character? I don't know... There are some archetypes that the AI often tries to show unwanted stereotypes no matter how much you try to say anything against it. For example, thanks to the stereotype of cyberpunk hackers being edgy, it once tried to give me a character who was a cyberpunk hacker that chose the name "Lucifer" for herself; this would usually be understandable, the problem is that, however, this specific one character was a "Christian girl" who happened to be a "cyberpunk hacker", and she'd quote the Bible while hacking demonic mega corps; but, like I said, some times the stereotypes scream through the algorithms.