So the fedi spammer is now just sending an image of a link to his discord instead of a wall of chinese glyphs.
This makes even server-wide post filtering ineffective.
The only real solution is to limit/block insufficiently administered servers that allow the abusive behavior (automated account registration and spamming).
OCR + other tricks could help in this specific instance.
The spammer has made the flaws in Mastodon's technical moderation capabilities quite apparent. Needs spamassassin.
@eriner this isn't as effective as ours. ours works on a user reputation system so an account has to meet specific criteria before it's passed through (user account age, follow/followings, post interaction)
Needs some form of scriptable reputational/scoring mechanism or something. Anything more than what it has now, which is effectively an all-or-nothing block and that's it.
@eriner that akkoma MRF for spam is like the dollar tree version of what we use but ours may be too heavy handed for some people. like for example if a user makes an account on your instance and they try to talk to us nobody will see it. they can still interact with us but until a threshold is met or >24 hours has passed nothing will be delivered to /inbox
@eriner@graf Maybe something like rspamd modified to parse inbound messages before passing them to the server. Would big a big project though since I don’t think most Fedi software is set up for that kind of filtering.
@istvan@eriner but we don't see any of it, thanks to the MRF. so why is something additional needed on top of that? mastodon and other development just needs to catch up to what we were doing a year ago
@istvan@eriner nevermind im retarded I see you were talking about what they can do to prevent it not what we can do to not see it. OCR is probably the best bet for the specific task which is something misskey has the ability to easily achieve as they already scan media for "NSFW" using AI.
we curbed this by limiting the total amount of times the registration endpoint can be hit in a minute, in an hour, in a day etc as well as limiting amounts per IP, user agent
Add comment