Side channels are the only way, practically speaking. Call them and have them read off the hash of their key and verify it against the key you received from "them". Or have them send a nonce value through the system to you that you provided over a side channel.

Any time you have a central authority providing that validation, you have to trust that your "friend" hasn't fooled them.

The theory with Signal is that you might verify their identity in meatspace, or some other way. But yes, it's definitely an interesting chicken and egg problem. As far as the thought exercise goes, it might be worth limiting your scope at this point.

Threema has a system where contacts are verified using a QR code scan: https://threema.ch/en/faq/levels_expl These three dots are shown throughout the UI, so with the identity of a contact you always see the trust level.

This is much easier to explain to a non-techie than comparing security numbers or hashes.

- Red = No trust

- Yellow = You trust that the server has verified the identity

- Green = You have verified the identity yourself

(Edit: Formatting)

Opinions about giving this trust matter to dedicated certificate authorities instead, like we do with HTTPS?

Edit: Found the problem with that. Certificate authorities are supposed to actually go out and confirm that the person who requested the certificate, is the person that they say they are. This is far too much work, if you want to verify each individual person who wants to use a messenger.