The “dead simple solution” is to just run the Signal protocol over SMTP, although I’m sure it’s possible there is a better design if you were to think about the specifics.
It's a much better user experience, and honestly I'm surprised that no enterprise orgs have adopted it, because it would probably be cheaper than all this phishing training.
1. At least until recent, Microsoft implemented it as blocking code in the UI thread – open a message and Outlook won't paint until it can verify the cert, access your local key store (hope your token is in a USB port which is 100% reliable), etc. If you thought “Does that mean that revocation checks block the UI until a network process completes?” you're sadly right.
2. Adoption hasn't been enough to be able to ban untrusted senders. This could still be quite useful for, say, a hard requirement that *@example.com must have signatures but it doesn't help with really common phishing tactics like pretending to be a vendor, business partner, etc.
3. You definitely need to get serious about key management because your ability to read your old encrypted email depends on retaining those keys. This is obviously not impossible but it has a cost when you think about the need to store them securely in a manner which can be restored after the inevitable system failures and compromises.
Preventing forgery of @example.com is nice, but sadly doesn't matter that much, because a message fro "Your Boss" <totallyfake@superscammy.example.org> is just going to show up as Your Boss in the UI with most clients these days, amd Enterprise oriented clients are worse than the norm. The fastest way to see the actual addresses is to just press reply, which is irritating.
From: "John Q Boss"
Subject: "the blog is down"
Can you check <a href="http://0-dayfuntimes.example.org">http://blog.example.com</a>
Yikes. I should probably be surprised, but I guess I'm not. This sort of thing goes a long ways towards explaining all sorts of other oddities or using Outlook.
They seem to be extremely hesitant to change it, which I assume is due to the corporate plugin ecosystem.
People use email because everyone uses email; you're not going to get people to change how they use email entirely. Nobody cares that it's running on SMTP, they care that they can email everyone they know with an email address.
Presumably we're discussing how an open protocol addition might gain any traction at all over walled garden protocols like Slack -- and in those cases you want to maintain as much compatibility as possible.
"Federated SecureEmail 2.0" would be dead-on-arrival, where "Secure Client on top of bog-standard email" is ever so slightly less DOA. You get to re-use the existing identity/routing system, existing servers, existing authorization, etc.
Why not just toss the whole shebang and rebuild it below that layer? Signal Protocol seems to be pretty successful here.
>Presumably we're discussing how an open protocol addition might gain any traction at all over walled garden protocols like Slack
No, I'm asking how a new open protocol can be built on top of email in a way that maintains strong backwards compatibility while offering strong security guarantees, like end-to-end encryption. I don't think it's possible.
You can't have strong security guarantees with backwards compatability, because backwards compatability requires plain text sending.
Unless you're OK with limited guarantees like the UI will tell you when the mail could be sent in plain text. Or if the UI tells you the message will be end to end encrypted, it won't fallback to plain text, etc.
That's right.
>Unless you're OK with limited guarantees like the UI will tell you when the mail could be sent in plain text. Or if the UI tells you the message will be end to end encrypted, it won't fallback to plain text, etc.
This would require changing all the clients.
Not really, the old clients would never send (or receive) end to end encrypted messages, so they don't need new UI to tell you that. That's the cost of unbounded backwards comptability.
If you want to have 100% coverage of email users, changing all clients is part of the deal. But, uhhh, good luck with that. Server based standards are an easier lift --- you could build a standard to require hop to hop encryption on mail and expect that to plausibly gain enough adoption to use over a managable time frame. Of course it would be trivial for a hop in the delivery path to subvert that and remove the request, and all hops in delivery would still have message content access; but if people like it, a large majority of servers might support it in 5-10 years.
You're describing TLS, and this is already happening.
Published November 2019. I didn't see any data on support, and hadn't heard of it before looking just now.
That's exactly what Autocrypt is doing. They're still at level 1, ie opportunistic encryption: try to encrypt if possible, default to plain-text if not. That's 100% backwards-compatible.
What we all want is a system where we are sure that messages can't be sent in plaintext. The only way this can happen is if MTAs actually read messages and check if encryption is applied; if not, reject it. That's something that can happen but by definition it can't be backwards-compatible.
I don't feel confident enough to actually try this project yet, but at least I wrote what my ideal chat app would be, oh and came up with the fancy name 'chattaur' for it.
1. Write the message.
2. Encrypt the message.
3. Paste the encrypted message into the email client.
Your odds of accidentally sending a message in the clear are zero.
As the number of people on the email chain approaches 2, the chance of someone accidentally copying+pasting the entire email chain decrypted into a reply reaches shockingly high levels.
A painful manual process where the simpler slightly-less painful path works, but results in insecurity happening, are not a good way to make sure security happens.
Heck, people (myself included) regularly mess up Reply and Reply All :-)
Still, doing so solves the transport and persistence problems you would otherwise have to deal with.
I'm pretty sure the dead simple solution is to either use SMTP or Signal and not a frankenstein's monster of both.