Consider this paragraph from the Signal protocol
> ...For example, they may compare public key fingerprints manually, or by scanning a QR code. Methods for doing this are outside the scope of this document.
> If authentication is not performed, the parties receive no cryptographic guarantee as to who they are communicating with.
Nobody I know in practice does this authentication. If an active attack is carried out, I'm reasonably sure people would not notice.
If the attack happened after you have started communicating wouldn't you get a notification that the key of those you're communicating with has changed? An attack would only be invisible if it happened from the first communication. But unless it's very targeted someone somewhere would compare the keys and notice.
This is the scenario I'm talking about.
That being said it's possible for all of Bob's contacts to have been MITM but not all of Carol's so each person does have to check at least one conversation to be sure.
Additionally, that ID+key would have to be updated somehow by the user and now you have a new problem: how do you verify that the user is who they say they are? Telling them to digitally sign something is possible, but what if the private key is gone?
The Signal protocol solves what it can solve for now. If cryptography experts have suggestions to improve it, I'm sure they can contribute.
Is there any protocol that solves the issue you're describing? And that has a good app? And that isn't centralized? We're asking for the "eierlegende Wollmilchsau" aka unicorn.
If they were eschewing some better option than "don't trust the server", that would be a problem, but is there a better option?
It's kind of fascinating, they have done a pretty good job of laying out their goals, easy contact discovery, forced protocol upgrades, etc, and people come up with all these improvements they could make by just setting aside those goals.
Indeed, EVERY decentralized communications system will have this problem. There's no way to establish trust over a compromised channel without either a centralized pre-trusted agent (like the CA system in TLS) or an out-of-band secure channel (like Signal, Matrix, PGP, etc).
Looks like the situation is more complicated and still unresolved for iOS: https://github.com/signalapp/Signal-iOS/issues/641
Meta-data: Log of who called whom, when and for how long. Also, user account details.