Sending an encrypted or signed message to someone who doesn't have your keys throws all kinds of scary messages in most email clients. Maybe not a problem for us™, but a huge problem for most people who find they can't send messages to their friends without remembering to do step X for people A, B, and D, but not C or Q. And even if you explained this to them, will they remember a month down the line when their friend calls them because they think their computer got hacked? (true story. nobody reads error messages, and encryption-related ones are among the most cryptic and scary looking.)
Then try convince them to keep using encryption / signatures for you, when it breaks for other people and makes sending an email more complicated, and they have to keep track of who C and Q are. Then try to convince them that, even though they have nothing to hide, encryption is still useful.
Then do this all over again when they get a new computer and forget to install PGP and have already lost their entire keychain.
--
The crypto is here, and yes, it has never been more accessible. The software using it is still garbage.
Seriously. The way PGP should work is: Whenever you send an email, it puts a header in the outgoing message with your public key. The header is not normally shown to the user and is ignored as unknown by non-compliant email clients. Then, whenever you send a message to anyone you've ever received an email from, you already have their public key, so your email client automatically encrypts the message before sending it, and their client decrypts any message it receives encrypted.
This is obviously ignoring a whole bunch of problems. What happens if your public key changes? What happens if the attacker sends an email from your address to the user? (Presumably in both cases the recipient will get a message complaining that the key has changed, and DKIM and the fact that your email server authenticates you will help with the second.) But here's the thing: Those problems don't happen normally. The average user doesn't encounter them in the first six months of using the software. They just install a client that supports the protocol and automatically get encryption for messages exchanged with anyone else using a supported client, without having to do anything special.
I think this is one of those "the perfect is the enemy of the good" scenarios. The people who want encrypted email want it to be secure against the NSA coordinating with AT&T and your email provider. Which would be great, if it didn't make the UX so terrible that no one uses it and causes everyone to default to no encryption. Do the above and you still have good security if you verify public key fingerprints manually, but it makes the process of encrypting your email as simple as installing the software, and if you don't verify keys then you're still safer against a large variety of attacks than the primary alternative of not using any encryption at all.
Without that, it's not just useless, it's detrimental to the system, as there could be lots of bogus keys accepted by people (imagine a virus that automatically generates and adds a PGP key to mail clients before sending to everyone in the address book, just to make it more likely to pass spam filters). Bogus keys in the web of trust would be a big problem.
In fact, if PGP/GPG were more popular, I imagine there would be the accompanying glut of horrible passwords used (or duplication from easily gleaned passwords), and pretty soon some virus would start automatically signing things it shouldn't on infected systems, and then the web of trust that the system relies on for third party verification wouldn't be so trustworthy.
It's not that I forgot that part, it's that that's the hard part. That's the reason PGP is hard to use: They try to make sure you do it securely. And you can't have some third party do that part for you without trusting them, and the whole idea is not to have to trust any third parties. What public servers are you going to use here? Does each email user have to run their own server? Unless you have a single central server, how do you know which server corresponds to which user?
Automating web of trust could be interesting though. Imagine you get an email from a new user that you've never received any email from before. There is some new P2P network where if you have someone's public key, you ask that user whether they know the new user's public key, and they send back a signed response (either "this is the key I have" or "I don't have a key", signed either way with the known user's public key). Then if all your friends who have the new user's key agree on what it is, it's probably right. If nobody has it, you get encouraged to verify it manually (i.e. in person). And if they don't all agree you get the nasty warnings about something fishy going on.
>imagine a virus that automatically generates and adds a PGP key to mail clients before sending to everyone in the address book, just to make it more likely to pass spam filters
That seems like a low-effectiveness method of sending spam, given that the public key is uniquely identifying and tied to a sender address, so once the spam filter realizes everyone is marking all those messages as spam it can just blacklist everything sent using that key. Also, how is it different from existing PGP other than that more people would be using it? If you've infected a machine with a virus you can do whatever you want to it. You could just write the spam directly to the user's inbox, or send it out from their own address and sign it with their actual key. Compromised machine = you're screwed.
>the accompanying glut of horrible passwords used (or duplication from easily gleaned passwords)
This isn't even necessary for a virus. The problem with viruses is that they can stay resident until you type your password and then it doesn't matter how hard the password was.
I'm certainly not going to argue with this, it's the basic gist of my original reply. :)
> Automating web of trust could be interesting though. Imagine you get an email from a new user that you've never received any email from before. There is some new P2P network where if you have someone's public key, you ask that user whether they know the new user's public key, and they send back a signed response (either "this is the key I have" or "I don't have a key", signed either way with the known user's public key). Then if all your friends who have the new user's key agree on what it is, it's probably right. If nobody has it, you get encouraged to verify it manually (i.e. in person). And if they don't all agree you get the nasty warnings about something fishy going on.
Exactly. This is similar to what I was envisioning when I was talking about confidence levels. Having different levels such as "I have personally verified (signed)" and "I know of and reasonably trust this key based on people I trust" and making that public in some manner would allow a slew of interesting techniques to verifying public keys to different assurance levels.
Come to think or it, it sounds like what we need is for a social network to adopt this. Google+ with it's real name requirements might make a good fit, but maybe real name isn't what we care about, maybe we just care about email. Alternatively, some alterations to diaspora might work out well (I know little about it other than it's a roll your own social network that I think can work as a node of a larger network).
> That seems like a low-effectiveness method of sending spam, given that the public key is uniquely identifying and tied to a sender address, so once the spam filter realizes everyone is marking all those messages as spam it can just blacklist everything sent using that key. Also, how is it different from existing PGP other than that more people would be using it? If you've infected a machine with a virus you can do whatever you want to it. You could just write the spam directly to the user's inbox, or send it out from their own address and sign it with their actual key. Compromised machine = you're screwed.
I'm imagining a virus that generates one on the infected system for the address the mail client is configured for. That could be a LOT of new keys.
The problem is the thousands or millions of bogus keys that start being sent from addresses that previously didn't have ANY key associated with them (or did, but not through that machine), clog the web of trust if they make it on there. If they are automatically added to mail client/PGP systems on the recipients end, that's a lot of bogus keys in users mail clients (even if it's just the 10% that arrive before spam filters react). If clients end up syncing their known keys to some central repo at some point, that's a LOT of bad data. I can imagine a case where someone generates a legitimate key and gets it personally signed by a few people, only to find that it's verified by hundreds of people on some public servers.
As for low-effectiveness, if it evades more filters by just a few percent, at the scales spam is sent that's a BIG deal.
> This isn't even necessary for a virus. The problem with viruses is that they can stay resident until you type your password and then it doesn't matter how hard the password was.
True. I imagine the really fast spreading and pervasive virus's need to be quicker than that though, but I have nothing other than a hunch to base that on.
That said, I'm still somewhat worried about MITM in the scenarios you describe above - should probably expose "you haven't verified this sender, it could be spoofed" somewhere, but somewhere icon/color-y rather than scary-error-message-y. Details, though...
EDIT:
Can't reply, but anyway, DNSSEC isn't really needed. That's just icing on the cake. What matters is a DNS record that specifies either the keyserver or the actual public keys themselves. A DNS TXT record or a custom PGP record will do just fine.
However, there's a huge training burden that needs to be overcome. People don't understand what it is, how it works, how to manage keys, what they can do securely or not, etc.
People complain about the tools, which I agree are insufficient right now for most people's general use—even with e.g. the very nice https://gpgtools.org —but I'm not convinced that's the primary hurdle.
Awareness of the technology is very low. I think it should be taught in public schools, and not as a specialized course but as part of general "life skills". I would love it if someone with a huge, visible platform like Google or Apple would push PGP integration into their tools for general use. Like, why isn't my PGP pubkey a normal field in my G+ profile? Or Facebook?
There are obvious answers there, of course, beyond the obscurity of the technology: if people start PGP-encrypting everything, those content platforms would lose value.
It would be like visiting Google's house to have a private conversation with a friend that Google couldn't hear. You can do it, but what do they get out of it anymore?
I'm not sure what the solutions are here, but I hope someone solves them.
What's needed is gpg in JavaScript, or a platform specific text input / gpg converter (you know a magnifying glass that hovers over encrypted text and shows the plain text)
Dunno if anyone has made one but till then we have no hope and Bob Hope
"OpenPGP encryption for Webmail"
Looks good though, with a dependence on browsers JavaScript vm it must be a nightmare of security
http://privacy-pc.com/news/changing-threats-to-privacy-moxie... https://www.youtube.com/watch?v=-JKqJ7gt5yk
Nothing that requires your average user to open a command prompt seems likely to be used outside of the most security conscious of circles.
HTTPS is less secure in practice (due to just being transport security and lots of intermediaries to attack), but still decent, and fairly widely used.
SSH is still the only cryptographic system which was so well implemented (in all ways, not just the cryptography) that it ended up taking over the entire market, displacing the non-secure options.
People should be building the next ssh, not the next PGP.
Yeah. I'd like something that simple and secure (where users can use the same private key across sites) for web browsers. X509 client authentication, as far as I can tell, doesn't cut the mustard. Among other problems, it requires trusted certificate authorities, which causes... problems.
It's really not an open crypto research problem; it's design and software engineering and entrepreneurship/marketing.
I killed it because in all that time, I received no encrypted messages. The only value I received was that it verified the signatures on a few mailing list posts.
I love the idea of ubiquitous public key encryption, but I think it just has a few too many moving pieces to get traction in it's current incarnation.
Maybe the NSA has it cracked, maybe not. But the IRS sure doesn't, nor do the FDA, DEA, FCC, FBI, Google, Apple, Facebook, Microsoft for that matter.
EDIT:
Here's how it works: I run PGP/GPG (GPG is the open source/free version of PGP, they're the same protocol) locally and the first time I use it, it creates two files, a "public key" and a "private key". I post/share my "public key" online (and you assume it's really me sharing it with you - this is important).
For example, here is my public key: http://neosmart.net/downloads/miscellania/mqudsi.ascAnyone can now use this key to send me a text message encrypted only for me. They just need this file and their own pgp/gpg private key; they tell it to encrypt message X for user with public key Y.
The result of that command (gpg --encrypt --recipient 'Mahmoud Al-Qudsi <[email protected]>' toencrypt.txt) will be encrypted text that no one but the holder of the private key (me) (which should never be shared) will be able to decode.
The result is plaintext. You can send it via email, text message, snail mail, whatever. There are apps that automate the encryption procedure as part of the sending process.