cross-posted from: https://lemmy.dbzer0.com/post/123857
This is my current attempt at preparing to counter the spam waves that will be appearing as the fediverse becomes more and more popular.
It involves the creation of whitelists based on a chain of trust between instances with easy ways to add and remove into it with few overheads.
Let me know what you think and if you’re interested, please do register your instance at https://overctrl.dbzer0.com.
I agree that we need far stronger admin and moderation tools to fight spam and bots. I disagree with the idea of a whitelist approach, and think taking even more from email (probably the largest federated system ever) could go a long way.
With email, there is no central authority granting “permission” for me to send stuff. There are technologies like SPF, DKIM, DMARC, and FcRDNS, which act as a minimum bar to reach before most servers trust you at all, then server-side spam filtering gets applied on top and happens at a user, domain, IP, and sometimes netblock level. When rejections occur, receiving servers provide rejection information, that let me figure out what is wrong and contact the admins of that particular server. (Establish a baseline of trust, punish if trust is violated)
A gray-listing system for new users or domains could generate reports once there is a sufficient amount of activity to ease the information gathering an admin would have to do in order to trust a certain domain. Additionally, I think establishing a way for admins to share their blacklisting actions regarding spam or other malicious behavior (with verifiable proof) could achieve similar outcomes to whitelisting without forcing every instance operator to buy in to a centralized (or one of a few centralized) authority on this. This would basically be an RBL (which admins could choose to use) for Lemmy. This could be very customizable and allow for network effects (“I trust X admin, apply any server block they make to my instance too” sort of stuff).
I think enhancements to Lemmy itself would also address help. Lemmy itself could provide a framework for filtering and report when an instance refuses a federated message with relevant information, allowing admins to make informed decisions (and see when there are potential problems on their instance). Also having ways to attach proof of bad behavior to federated bans at an instance level, and some way to federate bans (again with proof) from servers that aren’t a user’s home instance.
Finally, as far as I can tell everything following a “Web of Trust” model (basically what you are proposing) has struggled to gain widespread adoption. I have never been to a key signing party. I once made a few proofs on keybase, but that platform never really went anywhere. This doesn’t mean your solution won’t work, it just concerns me a little.
I expanded a bit more on some of how email tooling could be used within lemmy in [this comment(https://lemmy.nrd.li/comment/114218) as well. My ideas aren’t fully baked yet, but I hope they at least make some sense.
The problem with using email as a playbook is… Email has soft failed as a distributed system. It’s been captured by a few mega corporations who have created a web of trust between them and everyone else is struggling to get through and silently dropped in the spam bin often enough with no recourse. Cory doctorow has spoken extensively on this.
I think to avoid this happening to the fediverse as well we need to start building our own web of trust early on and not let the fate of email happen once more
Perhaps I am a unicorn, but I have self-hosted my email for years and don’t have deliverability problems. The only problems I have had:
To me one of the big things that differentiates Lemmy (and the fediverse in general) from email is that most of it is public, so the things in email that would involve sharing someone’s private information (email addresses, IPs, email contents, etc) are public (at least the post/comment and username+instance), and can all be verified. I think there is a lot of potential because of this. Maybe I’m crazy, but I just really don’t like the idea of a whitelist-based system because it means I as a small instance operator may have to sign up to dozens of services like the one you are building. I want my instance to be able to federate pretty much as widely as possible, and to me such a burden is too much to ask within a system/protocol/fediverse that is designed to facilitate sharing and decentralization.
Also, I think there is already room for a problem with “capture”. What motivation is there for .world .ml or beehaw to bother signing up for your thing? Even assuming you get 100 like minded admins to sign up for Overseer that is probably a pretty small fediverse island without them, some or all “mega” instances will probably just end up getting a pass anyways and at the end of the day no system is in place to help with the problem of bot/spamming users on trusted instances (whether in that WoT or just blindly trusted by the WoT).
Most of the spam I get is from gmail addresses, I don’t see it going any differently here.
Yes I am cognisant to the fact that it’s likely most won’t bother to sign. I’m considering just importing all instances and allow others to guarantee for them whether they’ve been claimed for not. Claiming am instance will likewise just allow their admins to guarantee and endorse other instances
Way I see it, they motivation to sign up is to crowdsourced trust building in the fediverse, instead of relying on defacto webs of trust which will develop organically around the big players.
About email, check the experience of doctorow https://doctorow.medium.com/dead-letters-73924aa19f9d