And you know another thing? Email works great to me, and it's decentralized. I have my email server, with my domain, so I don't depend on anyone to provide me a service. I have full control of my data that is on my server. I can even send encrypted emails with GnuPG without any problem, and it's as secure as Signal, if not better. I can use whatever client I want, a fancy application, a web interface, or as I do a small CLI program since I don't use graphical user interfaces.
Sure, having a chat protocol that is decentralized like email would be great! I wish that Matrix will evolve in something usable in the following years, we need that. I need a chat application where I can have a client that I can use in my terminal, and Signal doesn't do that (Telegram does, for example, since the API is more open, and it's what I use second to email).
Alas this requires a fairly contorted definition of "secure" to be true. The cryptography in GnuPG has plenty of problems even if you insist on manually doing everything from your own CLI tools which maybe don't suffer the problems from EFail.
Mostly it's just kinda old, AEAD hadn't been conceived when Phil wrote PGP you know. That's a big foundational idea for modern crypto and instead PGP had to kind of fudge a separate MAC into the design and hope that's good enough.
When you send your GnuPG encrypted mail, how do you decide which cryptographic primitives to use? With Signal the whole _point_ of Moxie's decision is that Signal gets to insist on all clients having whatever the best option is, so that's always used. But in GnuPG you've got to guess, what might the other participant's client be able to handle? If you guess wrong then your email is unreadable, or, worse, bad guys might be able to read it more easily.
> in most other European countries as far as I know
In England at least SIMs are available in the pound shop. You buy a SIM with cash (a single bright coin) and it comes with a phone number. British people do not carry any ID as a habit, and I make a point of never having ID unless I've been specifically asked to bring it for a good reason (e.g. to leave the country or when I got a job that required security clearance).
The only surprising thing that EFail revealed is that there are email clients out there that will silently allow html emails to communicate with the outside world. Encryption isn't the only thing that leaks from such clients.
>...PGP had to kind of fudge a separate MAC into the design and hope that's good enough.
Well isn't it? What attack is possible here? What attack was ever possible?
>When you send your GnuPG encrypted mail, how do you decide which cryptographic primitives to use?
You use the best ones supported by the receiver as listed in their public key. No one actually has to decide this and no guessing is required. This is something that the OpenPGP standard gets right for a crypto standard. Such things by necessity have to evolve over time. Sure this was hard to figure out, but it's done now. This particular wheel does not have to be reinvented or avoided.
No. Unsurprisingly the result is exactly what you'd expect. Idiots build software that throws away the error result and returns the unauthenticated text. This has always happened, which is why AEAD modes exist now. EFail documents what it names "CFB gadgets" to abuse this in typical HTML-based OpenPGP clients but you could attempt the same fun attacking a human subject directly, in some ways it might be easier because humans tend to just sort of "read past" nonsense in the search for meaning.
Phil Zimmerman didn't have a better option. You do.
> You use the best ones supported by the receiver as listed in their public key
So, you never use anything in the least bit new unless you're communicating with somebody who just minted new keys. For older users, you're stuck with whatever was current in the software version they ran five, ten, twenty years ago.
GPG2 doesn't support V3 keys anymore so old keys just won't work. That point is fairly moot in that pretty much all the ancient encryption is still unbroken.
OK, and how many non-technical people do you email that have GPG keys? If they don't have GPG, how do you have end to end encrypted communications with them?
This is where Signal won and GPG lost, with the Signal protocol integration into WhatsApp, one billion people instantly got secure comms without even having to know what a key is.
> it's as secure as Signal, if not better.
Signal uses a new key for every single message you send, so it could be argued that actually Signal is better than GPG.
Have you looked at signald or Signal-CLI?
https://gitlab.com/thefinn93/signald https://github.com/AsamK/signal-cli
When comparing to signal, or indeed to any modern chat client, you need to call out when you are mentioning non-Riot clients that don't support encryption.
In the original article you did this well: when you claimed "no fragmentation," the next sentence explained that the claim didn't apply to encryption.
In general, matrix feature-boasting about things it can do without encryption is confusing people. Most new users are coming to matrix expecting encryption, and they get it. But then matrix advocacy presentations proceed from the ridiculous assumption that open-source chat without encryption would be interesting as something to move onto, while hiding this assumption. Please stop doing this to preserve your credibility.
The core Matrix stuff has been usable for years, and in many ways you've had better metadata and censorship protection available than Signal - given you have the option of running your own server which could be entirely off the grid, and used only for your own conversations.
However, if your point is that Matrix focuses on freedom as well as privacy (rather than privacy at the expense of freedom), then we're agreed :D
When you say we're trying to improve I want to believe you. When you stubbornly declare "we're better" in "many ways" while actually being much worse, what could I conclude? Only that your goal is to deceive people because you're not really interested in improving. Blergh.
Back in February 2015 you did not understand the gravity of the mistake in leaving end-to-end encryption as a TODO item for Matrix and defended this decision as "Pragmatic" repeatedly. Matrix itself is a bit older, mid-2014 I think, but 2015 was the first time I ran into the arguments for Matrix. It's important to underscore those dates. I've moaned (including to his face) about Tim's choice not to do encryption out of the box in his toy hypermedia system, but that was almost 25 years earlier and whilst annoying a lot more understandable. 2014 on the other hand is after BCP #188 ("Pervasive Monitoring Is An Attack") and all that implies so there are no excuses. Matrix was a defective design on day zero. 2020 begins with a promise that it'll be fixed soon, kinda, sorta and then a retreat to old arguments about how it's all worth it for "freedom".
When I say "kinda sorta done" I am not referring to being able to send unprotected plaintext over the Internet in 2020. What's "kinda sorta done" is the basic task of sending messages between participants without anybody else knowing what was said, or preferably who by and who to. It sounds like Matrix's _ambition_ is still that in 2020 this will be the case for certain groups of participants, at least some of the time. Not everybody, and not all the time, and not yet unless you go out of your way which we know users will not do.
_Of course_ it will be difficult. But the same argument could be made for building a completely unencrypted messenger app. You won't have to worry about managing keys, obfuscating metadata, or anything like that.
At the end of the day, it's about realizing the importance of both privacy _and_ decentralization that our current infrastructure threatens.
And then he turns around and releases Signal-Desktop based on Electron?!
And two, "In Riot, Admins are shown in the membership list with a golden shield, and Moderators are shown with a silver shield. Other clients use similar metaphors."
If you actually used Matrix, you'd notice in whatever client that you're using that it's actually very easy to see who the mods are.
E2E on Matrix works, plus key verification is easier than on Signal. Managing metadata is hard, but my Matrix homeserver doesn't have my phone number (unlike Signal) and does not require Google Cloud Messaging. I can even run it on a PinePhone or Pocket CHIP!
>But for an actual user who needed privacy and security any time between then and now - and for future users who need it between now and whenever you get this stuff working in the real world, it was Signal that delivered. Moxie was right so far.
Tell that to the people getting imprisoned due to Signal's metadata leaks: https://news.ycombinator.com/item?id=21747424
Isn't this rather Twitter's metadata leaks in your source?
I don't use Signal, and won't as long as they force a phone number, but at least be accurate.
Please explain how Signal leaks metadata given:
- everything behind the client and the server goes over TLS
- all Signal messages are end to end encrypted with the Signal protocol
- the Signal server doesn't even know who is messaging whom in the majority of cases: https://signal.org/blog/sealed-sender/
In dev you also have libQuotient, metaolm and matrix-purple.
matrix-python-sdk also has support, but got replaced by matrix-nio.
They are both idealistic in their own way.
All these years later Matrix only has... The ambition to some day try to offer the core privacy features Signal already delivered back then. Some of the most basic stuff is, you believe, almost kinda sorta done.
This is, to be clear, much better than just sitting back insisting you were right but not lifting a finger. But for an actual user who needed privacy and security any time between then and now - and for future users who need it between now and whenever you get this stuff working in the real world, it was Signal that delivered. Moxie was right so far.