If a user/keyword/whatever gets negative karma because previously-blocked nazis fawn over it, newly posted Giorgia Meloni videos will carry a heavy burden, even if that particular video isn't anything to complain about. The ML will guess that the nazis are going to like it, and load the new videos with negative karma.
Thinking aloud, I suppose there are three kinds of bans:
1 - Actual, explicit bans (formal)
2 - Intentional shadowbans (decided by humans)
3 - De-facto shadowbans (unintentional artifacts of ML vertex weighting)
Note that the possibility of 3 occurs even if human intervention is needed to act upon the posts flagged by the ML --- the ML is, in effect, driving the 'attentional map' of the content moderator.
So this specific incident would probably be a case of #3, but it makes people angrier if you suggest it's #2, so that's what people are doing.