I mean, you suggest using AI to scan comments/code.
Maybe think harder and don't avoid a certain tech that is in a bubble and may soon be sued out of existence.
I do agree, github and other places that allow user generated content need to do a better job. I have more bots than humans following on twitter (and I have thousands of followers), for example, however trying to say "AI CAN FIX IT" won't win you any favors.
FYI I have a suggestion, maybe spin up said AI model on AWS and try scanning all the comments/issues github receives in a day. When even Microsoft can't pay your bill, you'll see why they do not yet use AI for that...though I'm sure they will try eventually.
On the other hand, I don't think posting the same spam nearly 1k times in 24 hours shouldn't raise some sort of alarm. 1 manual take down of a spam by a human should trigger a search (or have an option) to trigger a search through recent comments to see ones that are close. If I can search their whole site in 3 seconds they should have some sort of system that looks for 95% simular comments.
Seems simple but I am sure it's actually much much harder and has caveats and gotchas along the way as it scales.