Please provide links to some of these posts. We might be able to identify the specific reasons for that by looking at them.
It is true that we receive a significant amount of spam posts that are AI-generated. It is also true that when a topic receives a spam reply, this "bumps" the topic. The spam is typically removed within a matter of minutes, so by the time you visit the topic you might not be able to see any reason why it was bumped.
However, the volume of malicious AI-generated posts has not increased lately. And it is my impression that the volume of malicious posts has not increased since the advent of the AI era. The spammers just have a new way of generating the spam posts now. In my opinion, the switch to using AI-generated content for spam posts has actually been beneficial for moderation because this content has a very distinctive characteristic, which will immediately raise our suspicions. Previously, the spammers would usually copy/paste a relevant human written post from Arduino Forum or another platform like Stack Exchange (sometimes even doing things like automatically swapping words with synonyms to make the post content unique). Those posts were much more difficult to spot as malicious, both by the moderators and by the forum helpers. It was heartbreaking to find that one of these topics was only a vehicle for spam after the forum helpers had dedicated significant efforts to responding to a scumbag spammer.
As for "an AI", that is not really an accurate characterization of the situation, since the malicious posts surely originate from many different companies, not from a single AI. I'm not sure how much automation is used in their process. I wouldn't be surprised if it was still mostly human-driven (other than the generation of the text of the post).
I'm still here as always. It is true that we use an LLM to automatically detect spam posts:
However, although the specific Discourse AI system we are currently using is relatively new, it only replaces a different equivalent (but slightly inferior) system (Akismet) which was in use since 2021. We (the entire moderation team) still manually review every action taken by the system. So humans are still as essential in the management of the forum as has been the case since 2021. Prior to the migration to the Discourse forum framework in 2021, the management was a bit more human labor intensive due to the old forum framework's spam detection system being quite primitive (which meant far more false negatives which had to be managed entirely by humans), and due to the old moderation interface being a bit less efficient.
Yes. I did a careful search for the post. Even if it had been deleted, I would still have been able to see it due to having staff privileges.
Here is the discussion in question:
Occam's razor tells us that this was simply a case of a single human error (which is completely understandable), not some shady conspiracy theory. If a malicious entity had obtain the privileges necessary to delete posts by other users, why would they use those privileges to delete a single (or at least a small enough number that only one occurrence was noticed) random innocuous post, and thus risk exposing the breach before they had done anything meaningful with it? More likely they would either do their best to abuse the privileges covertly, or else make as large a disruption as was in their power all at once to take advantage of the small time window before such activity would result in the breach being closed and them losing the privilege.
There was a single occurrence of this months ago (a group of us discussed it in a direct message so check there if you want to refresh yourself on the details). It is true that this resulted in old discussions being bumped. However, that was months ago and there hasn't been anything else like that since.
Those bumps were a result of a large amount of manual work I did personally to preserve the valuable information that had been shared by other forum users on the affected topics. The forum software doesn't have any capability for doing such things automatically. So this is definitely something I would remember doing if it had occurred again recently (I'm still traumatized from the time I had to do it months ago 
).
Yes, this is another case (separate from the one @Grumpy_Mike) of when users might notice the side effects of forum maintenance. The one @sterretje is referring to was discussed here:
And another case of something similar here:
The forum maintainers periodically recategorize topics or fix breakage, even in old posts that we happen to stumble across, or receive a report about. However, there hasn't been any bulk campaigns such as the ones mentioned above recently, and the volume of sporadic maintenance hasn't increased recently.