This is how many spam filters have worked for a very long time. Using an LLM to help with classification could be a real option.

It wouldn't help with determining which security reports are slop and which are real, though.