An extended relative works at a digital forensic company where they are using LLMs to scan message bds and chats looking for child predators. Apparently their models are amazingly good. (the investigation, indictment still done by humans w/ subpoenas ofc, !minority.report yet)
If I had $fuck.u$, I'd hire them to write anti-racist/bigot/misogyny/phobic version that moderators could use. One click on a suspect account and it scrubs their post history for banvidence.