5 ms·
On the subject of encryption, S/MIME has broad client support (e.g. the clients shipped in all major desktop & mobile operating systems) and the security model
by acdha 9y ago
On the subject of encryption, S/MIME has broad client support (e.g. the clients shipped in all major desktop & mobile operating systems) and the security model is easier for most people to understand (WoT sadly often means “accept everything without checking” in practice).
What's missing is a LetsEncrypt-style effort to make it easy for people in non-managed environments to get and renew certificates. It'd be really nice if Mozilla & the rest of the industry chose to invest in that infrastructure and some usability improvements to make it more manageable.
This would be especially useful if it extended to browser integration so a user could trivially prove to a remote website that they had a valid signing certificate for a given address.
- rlpb 9y ago> WoT sadly often means “accept everything without checking” in practice I'd like to see Trust on First Use used automatically, combined with automatic key generation with no prompting. PGP is probably more suited to this. While it's not perfect, it is possible to retrospectively verify a key and on success consider it perfect. A well designed UI could expose this. Details: 1. Generate a key automatically with no prompting. 2. Sign everything outbound automatically with no prompting (if necessary use the header or rely on a multiple email handshake if you're bothered that users will see an "attachment" they won't understand). 3. Expose "identities" in the UI with information about what incoming mail corresponds to what identities, and what trust has been assigned to individual identities. 4. Permit encrypted email to be sent to identities, whether they are trusted or not. 5. Allow retrospective assignment of trust to identities. 6. Allow users to publicly "join" their identities for when autogeneration has happened in multiple places (eg. separately on webmail, my phone and my tablet). 7. Allow import and export of keys and external key agents for users who want to go all the way.
- tgsovlerkhgsel 9y agoS/MIME is extremely well suited for this, because it's already supported in many clients natively. You could make it possible to obtain a validated identity (e-mail validation) from a CA. Many of today's clients would natively be able to verify and trust this identity without previous key exchange. You could still slap arbitrary additional verification on top of this - trust-on-first-use, WoT, whatever you feel like - but an attacker would first have to overcome the (relatively weak) CA validation and then the additional fancy thing, while it would be usable for normal users immediately. The only open issue would be multiple identities and joining them. This could be fixed by sharing the key (i.e. if you generate a key on your phone, tablet and PC, you'd have to take each of the three keys and place a copy on the other two devices).
- jcranmer 9y ago> What's missing is a LetsEncrypt-style effort to make it easy for people in non-managed environments to get and renew certificates. This is a far harder problem than it looks. Let's Encrypt basically relies on a demonstration of control of a web server, and the resulting certificate is only as strong as that validation. For email, the way you can demonstrate control is in being able to read and send emails originating from an address... but the point of S/MIME is to secure against other people who can also do that. That's sort of the conundrum with email encryption. Your connections to and from the email server are encrypted (assuming you're not using an idiotic client, but if you care about encryption, this assumption is valid). Most traffic (~80%) between the major service providers at large is already encrypted. So the only people who can access your email in the first place are the people who can access your email account, namely you, anyone you give access to, and administrators on your email server. Any practical way of rolling out universal encryption is going to have to involve the latter party anyways, so there's little point to going for automatic universal encryption at the S/MIME-level: making the MX/MX links be SSL and banning non-SSL connections at the MUA level are much more effectual.
- acdha 9y ago> This is a far harder problem than it looks. Let's Encrypt basically relies on a demonstration of control of a web server, and the resulting certificate is only as strong as that validation. For email, the way you can demonstrate control is in being able to read and send emails originating from an address... but the point of S/MIME is to secure against other people who can also do that. The situation's a bit more nuanced than that: S/MIME protects you in the case where traffic can be read but not altered (or it would incur far more risk to do so) and most users state changes over time so there's more of a window to attack every message than to attack the account ownership validation, especially if that protocol is hardened. I see this as similar to the way many people are working to encourage end-to-end encryption because it reduces the amount of interesting data sitting around to be compromised on a server. Yes, at some point a hostile admin can do a lot but that doesn't mean that we can't harden against more realistic attacks.
- zokier 9y ago> For email, the way you can demonstrate control is in being able to read and send emails originating from an address... but the point of S/MIME is to secure against other people who can also do that. It's not exactly obvious how this situation is significantly different from the web side of things. > So the only people who can access your email in the first place are the people who can access your email account, namely you, anyone you give access to, and administrators on your email server What I think you are going after here is that shared/managed hosting style setups have only limited security, but that is equally true for web as it is for email.