I think that's backwards. You cannot assume that a "private" message is actually private unless end-to-end encryption is advertised. You still cannot safely assume so unless you've checked the code yourself. Privacy is never the default.

Hence the move to call things "direct" messages, rather than "private" as it used to be.

You cant even be sure when you have checked the source code of the project -- whos to say that is actually what is deployed?

Reproducible builds.

But how do you know that the build you verified yourself is the one that is running on their server? It would be amazing to solve this problem - are there any solutions?

End-to-end yousually means client to client. So all you need to verify is client code.

Intel SGX, if you trust Intel. But the amount of code that runs on SGX is limited so usually you'd run only critical parts there.

A practical example: https://signal.org/blog/private-contact-discovery/

Wow! That is an interesting issue.

I was thinking of having a third-party trusted services that compares the hash of the deployed application to the one they independently compiled themselves.

But the complexity is nontrivial and there is enough variations between the output of the same source code across different build environments that would make hashes useless.

Another possibility is having trusted compilers that would send link the source code to the build in a trusted repository.

You also need an agreement from the recipient to keep it private.

End-to-end encryption is for client-client communication.

When the sender and recipient are connected directly to the same system, there are no endpoints.

From sendbird:

"Customers can choose to further encrypt the messages prior to sending and after receiving for end- to-end encryption as needed (but this will cause moderation tools and search features to malfunction)."

Reddit may prefer to not provide end-to-end encryption due to reasons, say, keyword marketing.

I read this as saying that sendbird doesn't provide these features but you can use copy-paste to wedge them in if you're so moved.

https://help.sendbird.com/hc/en-us/articles/235854108-Are-yo...

I began work on a project for exactly this 2 months ago.

https://aurorabbit.github.io/emojicrypt.com/

It encrypts messages with a passphrase using scrypt, AES-GCM, and sha256 for an HMAC.

Each v1 protocol message includes a salt/iv, scrypt N&r, HMAC, and whether or not to lower-case the password. All messages encrypted with the proof-of-concept will be decrypt-able by all future clients, as there's room for 13 more protocol versions.

The code is in complete disarray, and I wouldn't want anyone to sift through it in it's current state[1], but the proof-of-concept is finished, and life came up, so it's been sitting since.

I have a list of features[2] I want to add in the issues tracker :)

[1] if you do want to sift through it don't miss https://github.com/aurorabbit/libemojicrypt. I separated the repos without following through on that reorganization.

[2] https://github.com/aurorabbit/emojicrypt.com/issues

What's the point of HMAC when you already authenticate with GCM?

More buzzwords.

It appears to me that I simply misread, thank you for your assistance. I've edited it now.

The user who posted that has a strong history of looking for, and loudly proclaiming, malice in many decisions reddit makes. They're not necessarily wrong, though, just slightly more deserving of fact-checking than the average reddit user.

Whoever made the call to implement this third-party service as their private messaging system made a terrible call.