If there’s anything I have learned about IP based blocking, it’s very unreliable. Especially in a NAT’d world.
Great you “shadowbanned” an IP, but you also impacted many other people and devices behind that public IP including the bad actor.
IPv6 is supposed to make NAT irrelevant but adoption is still very low despite IPv4 deprecated more than 2 decades ago.
And IMHO, NAT won the fight against IPv6 because it’s backward compatible.
The entire /64 will nearly always be a single ISP customer, not thousands of customers behind one address as it can be for IPv4. And you can start by banning the /64 and then widen the mask, say, 4 bits at a time if abusive traffic continues from an adjacent range. It's not that hard to automate this. Then the /48 gets blocked only if you see abusive traffic from multiple ranges within it, implying that the whole range is controlled by the attacker, or that ISP does nothing about abusive customers, which is nearly the same thing.
In this scenario it doesn't matter. Some user might be able to access the malware still, but that's better than not blocking it at all.
Still an interesting idea though.
> Shadowbanning is extremely hostile to users that have been mis-identified as spammers (which will happen)
It should always be a manual action and moderators should continue to see messages of shadowbanned users. You can always lift it in case of a mistake.
If you're going to have a free tier on your service and your service has any sort of interaction going on between users that could be degraded by spammers and the mentally insane, you're going to need shadowbanning. It's either shadowbanning or upping the hurdle to creating an account considerably.
The risk of misidentifying legit users and shadowbanning them outweighs the potential gain.
What's the risk?
Because if done correctly the user never knows they are shadow-banned. It sounds trivial when you know _how_ the shadowban is done. But for instance, instead of an IP check, perhaps it's a time check - after 3 days it comes into play. Or a combination of different checks. So imagine that you are accessing a service that appears to be working correctly .... you would basically need to a) determine that that service even does shadowbanning, and b) think of infinite ways that you might be shadowbanned and try to determine if that's the case.
The second most effective thing is making the malicious actor use some sort of resource. Such as a payment (the author uses), or a time commitment (eg new accounts can only create 1 link a day), or some other source of friction. The idea is that for legitimate users the friction is acceptably low, but for consistent spammers the cost becomes too high.
The 3rd thing I've found effective is that lots of spam comes from robots - or perhaps robots farming tasks to humans. If you can determine how the traffic is coming in and then filter that traffic effectively without indicating failure, robots can happily spam away and you can happily filter away.