Hello. I am looking for an alternative to Telegram and I prefer an application that uses decentralised servers. My question is: why is the xmpp+omemo protocol not recommended on websites when it is open source and decentralised? The privacyguides.org website does not list xmpp+omemo as a recommended messaging service. Nor does this website include it in its comparison of private messaging services.
https://www.privacyguides.org/en/assets/img/cover/real-time-communication.webp
Why do you think xmpp and its messaging clients such as Conversations, Movim, Gajim, etc. do not appear in these guides?
It isn’t a meme. It is a fact of modern cryptography in many settings. For example TLS, which is a huge bulk of the traffic, guarantees again privacy not anonymity. I’m not saying one shouldn’t care about metadata privacy. Every communication one engages in requires risk benefit analysis. If your threat modeling shows that for a given message, anonymity is required, then signal, and nearly every single other protocol out there is insufficient.
That doesn’t mean TLS or lib signal, or any other cryptographic tool is not useful, especially in conjunction with other tools.
There are many cases where I want my messages to be private and the cost of entry for the message receiver to be low. Signal is great for that. But I’m not saying no other tools should be considered, just that signal is good at what it does.
“Anonymity” is a vague term which you introduced to this discussion; I’m talking about metadata privacy which is a much clearer concept.
TLS cannot prevent an observer from seeing the source and destination IPs, but it does include some actually-useful metadata mitigations such as Encrypted Client Hello, which encrypts (among other things) the Server Name Indicator. ECH a very mild mitigation, since the source and destination IPs are intrinsically out of scope for protection by TLS, but unlike Sealed Sender it is not an entirely theatrical use of cryptography: it does actually prevent an on-path observer from learning the server hostname (at least, if used alongside some DNS privacy system).
The on path part is also an important detail here: the entire world’s encrypted TLS traffic is not observable from a single choke point the way that the entire world’s Signal traffic is.
Don’t mistake me for saying you’re wrong. I agree with you, mostly. But sealed sender isn’t theater, in my view. It is a best effort attempt to mitigate one potential threat. I think everybody would like that solved but actually solving it isn’t easy as I understand it. Maybe not intractable, but if you have a solution, you can implement it. That is one of the things I like about free software.
In any case, I’m only saying Signal is good for a subset of privacy concerns. Certainly not that it is the best solution in all cases.
But, what is the potential threat which is mitigated by sealed sender? Can you describe a specific attack scenario (eg, what are the attacker’s goals, and what capabilities do you assume the attacker has) which would be possible if Signal didn’t have sealed sender but which is no longer possible because sealed sender exists?
Sure. If a state serves a subpoena to gather logs for metadata analysis, sealed sender will prevent associating senders to receivers, making this task very difficult.
On the other hand, what it doesn’t address is if the host itself is compromised where sealed sender can be disabled allowing such analysis (not posthoc though). This is also probably sensitive to strong actors with sufficient resources via a timing attack.
But still, as long as the server is accepting sealed sender messages the mitigation is useful.
Pre sealed-sender they already claimed not to keep metadata logs, so, complying with such a subpoena[1] should already have required them to change the behavior of their server software.
If a state wanted to order them to add metadata logging in a non-sealed-sender world, wouldn’t they also probably ask them to log IPs for all client-server interactions (which would enable breaking sealed-sender through a trivial correlation)?
Note that defeating sealed sender doesn’t require any kind of high-resolution timing or costly analysis; with an adversary-controlled server (eg, one where a state adversary has compelled the operator to alter the server’s behavior via a National Security Letter or something) it is easy to simply record the IP which sent each “sealed” message and also record which account(s) are checked from which IPs at all times.
it would more likely be an NSL or some other legal instrument rather than a subpoena ↩︎
Yes, subpoena was poorly worded. NSL is more likely. But still it is a time-forward threat, which means there is value while the server is or was accepting sealed sender.
And I wasn’t suggesting timing attack is required to defeat sealed sender. I was, on the contrary, pointing out that was a threat even with sealed sender. Though that is non-trivial, especially with CGNAT.
So in summary. You’re right. Sealed sender is not a great solution. But it is a mitigation for the period where those messages are being accepted. A better solution is probably out there. I hope somebody implements it. In the meantime, for somebody who needs that level of metadata privacy, Signal isn’t the solution; maybe cwtch or briar.