There's also an old phrase that putting locks on your doors doesn't actually stop a determined attacker, but that it's okay because they're not meant to—that they're meant to "keep honest people honest". It's a principle that applies here.
Perhaps even functional regression: what, are you going to run a hash blocklist across all nodes? Like spamhaus? Is there logging or user accounting? Is anything chain of custody admissable, or are we actually talking about privacy and liberty here?
Is everything just marked, "not for unlimited distribution"? And we dwpend upon there not being bad actors?
Real costs are very different with just friendly early adopters.
Cryptographically signing posts (with LD-Signatures) may help with integrity, but that can be done with centralized systems and does nothing to help with confidentiality.
What about availability? Is it a trivially-DOS'able system?
Does it prevent screenshots and OCR?
So it's in standard record structs and that makes it harder for the bad guys?
Who moderates mean memes with my face on them?
It is my hope that future Linked Data spec tutorials model something benign like shapes or cells instead of people: so that we can still see the value.