Post

Yehh

As I've said before, deniability in cryptography matters little while you're actively using it - it becomes important when a 10 - 20 year old log is pulled from an old archive out of context.

ie most people i know talking about anything important are on signal here, and many places whatsapp, in various smallish overlapping groups plus some slacks and discords

not deniability but weak privacy and definitely strong pseudonymity friction is not security but it does change the shape

I also recommend a possibly controversial rename to “private data”. Much easier to understand and spell-check :)

It might be possible, but it’s definitely not what the standards as written want me to do

Like yeah, deniability is a smell. But is it as bad a smell as “let’s build a completely solution to solve it”? No. Not at all. Having a totally unrelated answer for private data is a massive smell.

Yeah honestly I just cannot get myself to care about that part. Am I right in thinking that you care about it because that’s what’s driving the wildly different architecture? Because yeah, in that case I definitely want the requirement to go away.

We'll see if it extends to more than just this one rant!

i don't think users expect that things they post are signed by their pds in a way where they're hard to deny after the fact, as one could with doctored screenshots. it's kinda bad enough in public atproto, but it's needed for relaying records. if we don't have that requirement anymore, why keep it?

AT added that capability to help build new models for distributing public content, on the premise that this "equal playing field" tradeoff -- everybody has the same access/proof -- is better than existing models But that premise may not play well with folks with more serious concerns about privacy

Normal, non-deniable crypto provides two features here: 1. It keeps things simple. Better for builders, easier to understand. “Everything works the same except for the ACLs” 2. It enables redistribution in positive cases, if we can design redistribution systems that maintain the ACLs.

besides maintaining the user's expectation of privacy in a permissioned space, deniability is just a very distinct, understandable difference from public data. without it, users may not trust the stated benefits of using permissioned spaces, since they don't understand any of the other differences

I think you’re misunderstanding what deniability does. Content isn’t any more “non-existent” because it’s not deniable. The only question is what happens *after* it leaks, not how easy it is to find or leak.

For which use cases is deniability critical?

My lukewarm take is that everything will end up being permissioned data unless a whole lot of energy and effort is put to prevent that from happening. I think the public square is important and needs to be protected.

Thanks for the write up - I copied it into the forum and asked you a question about patterns of dev inside / outside the IETF over time discourse.atprotocol.community/t/all-data-s...