Sarah Jamie LewisJan 5

RE: https://chaos.social/@delta/115842446685941927

Encrypted content / headers / transport / etc. are great harm reduction, but by no means is it metadata avoidance (in the strict academic sense of what those words mean when applied to communications systems).

Third party actors (e.g. relay servers / network actors) can build correlative models using the *metadata* inherent in the communication protocol (source, destination, timing, frequency).

We have decades of research demonstrating the power of such models - let's not minimize them.

Show threadSarah Jamie LewisJan 5

Resisting such correlative models is very much still an active research that spans anonymous communication networks / mixnets / application security and many other subfields.

It's something I've been working on for over a decade, and while I am very proud of the work we've done on @cwtch there is still a long way to go (if you want to dig into what we consider the security model here you can check out our security handbook https://docs.cwtch.im/security/risk)

Risk Model | Cwtch

Communications metadata is known to be exploited by various adversaries to

Show threadSarah Jamie LewisJan 5

Right now I believe that tor v3 onion services are the closest we have to practical approaches to minimize metadata (far from perfect, and the price point for defeating it gets lower)

There has also been a lot of really cool research in the last few years specifically on new cryptographic protocols in aid of solving this problem in a way that is different from the "we need a verifiable mixnet / PIR" approach see keywords: fuzzy message detection / oblivious message detection / OMR

Show threadSarah Jamie LewisJan 5

My personal long term take here is that a good solution for this problem might look something Niwl ( https://git.openprivacy.ca/openprivacy/niwl ) which is a prototype I built some years ago based on playing with FMD/FuzzyTags

i.e. how can we approximate a mixnet/PIR system with untrusted actors in a way that minimizes communication bandwidth?

niwl

a prototype system for open, decentralized, metadata resistant communication using fuzzytags and random ejection mixers

Open Privacy Gitea
Show threadSarah Jamie LewisJan 5

As that work says, and what I am slowly convincing myself should be emblazoned across every project that even remotely attempts to make progress on communications privacy: hic sunt dracones

(But that isn't good marketing, so...)

Show threadSarah Jamie LewisJan 5

And since it's come up:

If you are willing to trust a routing intermediary (and anyone who may force that intermediary to act e.g. a government) - then you don't have to worry about such attacks.

But in that case, just use Signal - it has a lot of other nice cryptographic properties that apply in that model.

(Signal also make various claims about metadata, I disagree with many of those claims, some I find outright ethically dubious as a privacy researcher - but you should just use Signal)

Show threadSarah Jamie LewisJan 5
(For those who care: Signal used to place a heavier emphasis on trusted execution environments as a mechanism for protecting metadata, despite TEE's having been incredibly broken in most realistic threat setups forever, in recent years Signal switched to a "we don't collect anything" messaging which is fine for what Signal is, but I do wish they would be more honest about what they *could* be compelled to collect in theory. (which in reality is: a lot of metadata))
Show threadSarah Jamie Lewis
The caveat to all the above being that I haven't deeply looked into Signal for a few years since I reported a UI redressing security issue to them and they fixed it without acknowledging the fix in the release notes and personally I find that a little not great ethics wise.
Jan 5 at 9:18pmWeb
Show threadSarah Jamie LewisJan 5

To close: I only want to bother with critiques for projects I actually think have the capacity to do better and make the world a better place.

The problems in this space are so large and multifaceted that we need people going out and exploring different approaches.

But when it comes to privacy tooling, being honest and upfront with people about the nuance matters - and my goal with the above thread is to bring some of that nuance.