over the last maybe year or so we have been keeping registrations closed whether it be for abuse reasons or because of election results, we have used closing to our advantage to prevent illegal activities on poast.
over the last year or so, pedophiles from pawoo, baraag wherever have been coming to poast because of how well we are federated and searching for pedophillic terms, images or likeminded individuals using hashtags, usernames etc.
i spent the entirety of yesterday and tonight and will spend a couple days going forward writing a python program that analyzes our logs in real time looking for these behaviors. identifies these threat actors and emails staff with details including their account identities if they have one.
in the next two to three weeks i will be testing this fully in production where any of these actors seeking this material will have their information emailed to us in real time. if it passes my thresholds for account age, reputation etc and there are no false positives we will begin automatically formatting professional emails with reports of these users to NCMEC and cypertip.ca. people outside of north america unfortunately have to be dealt with through NCMEC but it is a step in the right direction.
I am focused on finishing this and I will make this software public so any other instance can also utilize the service. pedophilia has no place on fediverse just like indians and trans people.
@graf co-admin on here had a hacky awk script which did much the same (except for the ncmec thing) and it worked pretty well for a while, only problem we had was when nginx's access.log would rotate and that would totally fuck up the script, looking back the script probably still had a stale fd opened after rotation that would never have anything come down
@nyanide yeah, mine accounts for log rotation already. first thing I thought of. will keep you in mind when I'm done. will make auto reporting optional as well and will try to only use included python libraries so no need to install 3rd party trannyware
@graf Doing gods work here! Good to see that your on top of those disgusting freaks, First thing i did was defed with baraag almost instantly because of the kind of shit that goes on there.
> only problem we had was when nginx's access.log would rotate
Couple of solutions to that. I changed the rotation script to copy/truncate, but you can also just have it restart periodically.
> the script probably still had a stale fd opened after rotation that would never have anything come down
That is it, yeah. It's not even a stale FD, it's just an FD that points at a specific file, and the file has been moved: it's still reading that file, just the file is no longer the one getting new lines because logrotate kicks demons when it's done to have them reopen the files.
>from pawoo, baraag wherever I didn't know baraag had actual pedos, I thought it was just western loli artists who (almost) all draw the ugliest fucking shit. pawoo is expected though, given that they've actually clamped down a lot (though I have seen a pedo on pawoo local TL in the last week, so seems like they're not catching 100% during registration). As much as I'm happy for it being cleaned up, this is the unfortunate consequence.
@nyanide im really bad at commenting shit so I will go through this weekend and comment everything for you bro so you don't have to hunt. I will write documentation for it as well but most of us comments are better than docs 100% of the time
@graf Yeah looking through the code is probably the first thing I'm gonna do when you put it out there. Definitely looking forward to giving back if there's anything I feel that might need improvement.
@graf@nyanide docstrings man, it is possible to "auto" generate documentation from the docstrings, you just need to write good docstrings, preferably type annotated and when using **kwargs tables writing the list of keywords and the type each one expects.