Keybase's New Key Model

http://i.imgur.com/R9f96Wr.png

(get it, it's a key-bass ... whew)

The most important benefit is probably that NaCL keys are 256-bit keys based on ECC (specifically, Curve25519), which means these keys are stronger, faster, and actually more secure than the strongest keys commonly used in PGP (4096-bit RSA keys). Critically, they are much smaller and therefore easier to communicate between people/devices.

While their ideal workflow is to communicate keys by scanning QR codes, they currently need an alternative workflow to communicate keys between desktops (and will always need an alternative for users who can't scan QR codes for whatever reason). The "Paper keys" section of the blog post discusses this at length, and shows an example of a Paper Key:

"death punch correct staple battery horse clearly cherry picked words yeah moo car lisp"

They are probably generating this (Diceware-esque) string by splitting the 256-bit key into n-bit segments, using each n-bit segment as an index into an array of words, and then joining the words to form a "sentence" that encodes the value of the key.

14 words is already pretty unwieldy, as they acknowledge in the blog post. Imagine how bad it would be if you had to encode 4096 bits instead of just 256!

That was probably the most important motivation for Keybase to use them. Other motivations may have included:

* ECC is generally considered more future-proof than RSA

* NaCL is a small library and therefore is probably easier to incorporate into a variety of clients rather than the behemoth that is GPG/libgcrypt.

* NaCL is the new hotness, PGP is old and busted

There's a document describing this client/service split here: https://keybase.io/docs/client/client_architecture

This lets me use my key on an Android device (via Open Keychain), or PC (via GnuPG's smart card interface.)

I don't want additional NaCL keys hanging around on 'online' machines outside of a secure element such as a smart card.

As best I can gather, this ties signing to keybase.io (which is anchored in a GPG-key). But it's unclear to me how this is an improvement for users. If you sign anything with a nacl sub-key, that can't be verified with gpg? It's not even clear if this will work off-line, when keybase is down, when keybase have revoked your account...?

> Technically, though, this is not a normal 2fa: the devices will share a secret with just each other.

This is exactly 2fa: two devices sharing a secret? Normally your idp and your TOTP app/token share a secret, now two of your devices share a secret?

I very much welcome someone hammering out a standard equivalent to GPG, but based on NaCL, that is: have a master key (off-line), have a private-sub-key (on-line, protected by smart card/pass phrase), enable further sub-keys -- and a companion tool (set) for verifying up the chain (like gpg2 --verify file.asc, but with file.asc being backed by a (device) sub-key, signed by the private key, signed by the "master" off-line key -- and allow for revoking "from this key and down" (eg: I lost my phone, revoke phone-key, my laptop got hacked, revoke my sub-key, and all device keys...)).

But I'm not sure if that is what keybase is trying to do?

I'm not even sure if it really makes sense to keep backwards comparability with GPG -- the main thing would probably be to be compatible with GPG agent and smartcards. For most everything else, I think a new standard would be a good thing. Perhaps with some kind of bridge, for easier transition (eg: nacl-gpg --verify file.asc, which gives similar output/return-codes to gpg, while file.asc can be signed by a "nacl-key" trusted by a gpg key. But the complexity would probably make for a less secure system until everyone could just move on to the new standard).

By the way, libsodium has the same maintainer than dnscrypt (and Pure-FTPd) but he does not work for OpenDNS anymore (afaik he works for OVH, but libsodium is his personal project).

To be clear, you'll be using this system automatically if you install Keybase on your phone or desktop. It'll just work, and you'll have a private key on that device. As you add devices, you'll collect keys. If you remove a device, that public key will no longer be a part of your identity.

If you want to use PGP on one of those devices, you can. And you can strongly connect it to your other keys, as the graphs in the post show.

This should be easier. From a "regular" non-PGP user perspective, you needn't think of this as managing keys, or adopting a more advanced key system. Instead you would just be managing your devices and installing Keybase.

That's our hope anyway. (Disclosure: I'm one of the people working on the project - https://keybase.io/chris - thanks for the kind words about the project)

I was waiting on smartcard support in that project myself

Google cache:

http://webcache.googleusercontent.com/search?q=cache:vneH1nu...

Historically, a few people using PGP would do this identity mapping, at least in one direction, by posting a PGP fingerprint (of their public key) on their, say, Twitter profile . Of course, this only went in one direction. To achieve a bidirectional proof, they would also have to post a signed statement going in the other direction, too, stating they were in fact that twitter user. (Otherwise you couldn't start with a message and conclude which Twitter user sent it - multiple people could claim this.)

Keybase started as a hobby project to make these posts structured and formal for PGP users. We just wanted to make this kind of thing easier, so we made a directory to hold the signatures going in the other direction, and a reference client that would look up these announcements and verify them.

The other specific advancement is that these signatures have been put into a data structure designed to prevent the server from lying by omission or forking - a big problem that exists with historic key servers. If you remove a key, what's preventing the server from not admitting that? Or what's preventing the server from giving Alice and Bob a difference experience from Charlie and Diane?

Anyway, that was the first idea behind Keybase. To make a better directory and walk people through these proofs.

What we realized - and what has gotten us both (a) extremely smart developers to join us, and (b) funding - is that there are deeper problems beyond the identity establishment....problems that can be solved in 2015.

Those two problems are (1) pretty sucky clients for public key crypto (at least from the perspective of non-programmers), and (2) key management issues. The first is something we have experience with, as our team has worked on a lot of successful projects. And let's be clear, lots of teams could solve this, if given the resources. The second is something we feel can finally be solved thanks to mobile phones. Phones and watches and other small devices allow you to provision new keys easily, because 2 devices can be brought together. That wasn't easy in the past. All this is now possible without understanding what a key is. Or worrying about how to store and move one around safely. That's the new goal: actual crypto clients. For signing, verifying, encrypting, decrypting, and sharing.

You can read the front page of keybase.io for some more info, but it's somewhat out of date and talks mostly about PGP. This will change when our clients are further along and we're seeking more beta testers.