It's using AI to filter or reduce interactions with words puritans and Chinese censors find "offensive" but to the other responses point, it goes deeper. They train the AI not to cuss, not to be able to discuss sex at all, and to avoid topics that the developers' companies or countries don't like. But we're never shown what they push for or are guarded from by the companies releasing them, so the shift is subtle.

The underlying tech doesn't care about swear words. But, from what I hear, the "AI" models and interfaces that are being marketed for general use have been intentionally trained not to emit such words, and may even be blinded to those words by their creators. I haven't verified it, but supposedly adding "fucking" to your Google / Bing search query will completely disable their "AI" generated results, e.g.