On 30 Jul 2022, at 18:26, Jeremie Miller <[email protected]> wrote:
> 
> Isn’t this somewhat overstating the likely privacy benefits? If the prover 
> reveals _any_ PII to the verifier then the verifier can collaborate with the 
> issuer to discover everything about that user.
> 
> JWP as a container aims to make unlinkability _possible_ for applications to 
> build, not a guarantee.  There are many extremes an application may choose to 
> design for to accomplish different scales of unlinkability (from multiple 
> verifiers colluding, from the verifier and issuer colluding, from multiple 
> presentations to the same verifier, etc).
> 
> In my mind it's akin to you can cryptographically validate the contents and 
> signature in a JWS, but how you decide if you trust the signer is up to the 
> application or higher level protocols.
> 
> And we know from many studies on deanonymisation that it is very easy to 
> accidentally reveal enough information to be identifiable. ZK proofs are nice 
> and everything but they only ensure zero *additional* knowledge is gained by 
> the verifier. In practice what is explicitly revealed is often enough. 
> 
> That's exactly why we believe this work is very important, having a container 
> to support algorithms where zero *additional* knowledge is revealed by the 
> container and crypto layers.
> 
> It *is* very easy to incidentally reveal linkable factors, which is why JWP 
> is hard to get right, and critical to do so to enable this capability.
> 
> IMO if you want to have any hope of actually achieving the privacy you want 
> then you really need to design the entire protocol, including specifying 
> exactly what information is to be revealed. I think designing a generic 
> “privacy preserving” message container is likely to give people unrealistic 
> expectations. 
> 
> We have the lowest level privacy algorithms becoming well established like 
> BBS (and CL signatures, etc), next we need a privacy-capable container to 
> make those algorithms more accessible and interoperable, then we need privacy 
> protocols to leverage those containers, then privacy aware applications, 
> ecosystems, and user experiences. 

I think perhaps we most fundamentally disagree on this roadmap. Although many 
standardised systems have followed this kind of modular design, I don’t think 
it is the best approach. Compare say IPSec vs WireGuard for an often-cited 
example. Privacy is not a separable concern. Start with the privacy-aware 
applications. Otherwise I think we’ll end up with lots of “privacy-related” 
tools with lots of sharp edges that inevitably get used inappropriately.

— Neil

_______________________________________________
jose mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/jose

Reply via email to