Keybase launches encrypted Git

I may be wrong, but here's my current understanding.

I believe Git CAN check the validity of sha1 hashes (I read the source a few years ago and have a very tiny git commit) using git fsck, which I believe kernel.org does nightly. It just doesn't do so automatically with every commit or whatever. But you can set up a test in your server, I believe, if that's important to you, either watching the files, or checking pushes which I believe github does. So that's not the issue.

It's sha-1 collision attacks that are a theoretical issue.

My understanding of the currently known SHA-1 attack is that it requires binary data (hence PDF files for the example) and requires you to control both the original file and the subsequent file. So an attack would have to generate an apparently innocent file and a malicious file both of which have a binary block, insert the innocent file into the repo, and then somehow, most likely outside of a git push given mitigations like github's, replace that innocent file with the malicious file.

Now to your question, checking in the PDF files from the proof of the attack in git doesn't work, because git also adds header info. And generating the files requires ~ $100,000 dollars worth of ec2 time, or the equivalent, so nobody has gone through the trouble of generating files that allow this specifically to prove it for git. Bit it's definitely possible, and cheap enough for a criminal organization or a state agency to do. Just because someone hasn't done it for git specifically shouldn't mean that the attack isn't possible, just that security researchers don't have unlimited funds, and the existing proof, while not specific to git shows the issue generally applies.

Last I saw, the git mailing list was debating sha3-256 and BLAKE vs SHA-256. There's some indication that SHA-256 may get intel HW support, and that may be useful for speed with really really big git repos (like microsoft's apparently). SHA-256 doesn't have an attack on it that's known but unlike ShA3-256 (and I believe BLAKE since it's a stream cipher) SHA-256 is a block cipher, so it's not stateful. That means, while no known attack exists, theoretically if an attack existed you could corrupt a specific block in a similar manner to SHA-1. But SHA-256 has been much more extensively tested for issues while SHA3-256 is newer... it was created ostensibly as a backup in case the current known safe standard of crypto like SHA-256 is attackable.

There are some issues with SHA-256 being used in repos that have signed SHA-1 hashes already, in terms of mapping SHA-256 to SHA-1 hashes without borking the signing. Obviously if you change the underlying structure of signed stuff to store a new hash, it changes the hash.

My personal thought would be to implement SHA-256 and SHA3-256 as options simultaneously, as they are both NIST standards, make SHA-256 the standard so big repos can be as fast as possible.

I am not a crypto expert, or a git expert though, so if I'm wrong, please correct me. Being wrong means I get to learn stuff and that's great!

Because anything that can be misused will be.

I’m sure there’s a law with someone’s name that states that. But just in case it hasn’t been claimed yet, I’m proposing that we call it the fuck you law. Because the next time someone comes to me to ask me to fix their trello to zappier to email to google sheets setup they use as a project management tool, I want to be able to say, “Fuck you and there’s a law that says so.”

Throwing my "I It seems to me that there's a lot of product opportunities in the corporate world that go beyond what Keybase is providing today. Chat and Git are interesting, but there's already a lot of momentum in both these areas. Been thinking how I use encryption and where things fall short today. One of those areas is build signing and hardware key management for our team.

Everything that goes on our servers get signed by an official PGP key. Only a couple people can sign builds, and each has a Yubikey with PGP subkeys on it. This is kind of annoying to manage. We use an airgapped computer that houses the private key, can create subkeys and assign to Yubikeys, can handle expiration management, etc. When we want to deal with this, we have to get the computer, unlock access, and deal with the command line. This is error-prone and annoying. Having a solution that allows for safe storage of a private key and easy management of subkeys on smartcards would be amazing without the need for an airgapped computer and a command line would be really interesting.

(The signing/verification part can probably be handled today by the keybase tool.)

Okay, that's maybe more specialized. Let's move away from paranoid server builds and go toward something similar that's gotten plenty of companies in trouble: Malicious e-mails. How often are we hearing about some poor employee receiving an e-mail that appears to come from a co-worker that contains a finance document with a trojan? Or maybe just a simple document with a form, instructions, and a link that results in information leaked to some third-party?

If there was a dead-simple way to sign and validate documents over Keybase (and I mean dead-simple, built for people who only know Word and Excel), for use in e-mail and document management, with marketing around "For $XX/user/month, you don't have to worry about getting hacked," I bet plenty of companies would bite.

I don't know what that looks like exactly, but just playing around loosely with some thoughts, it would be interesting (particularly for fully IT-managed systems) to have a Keybase Shield product that would automate much of the signing and verification of documents. It could tie into Word, Excel, etc. via their plugin interface and sign on save, and/or provide a big "Sign this document" widget on the side of the screen that a document can be dropped onto (or a Share action on phones). It'd then own the file associations for these documents, intercepting them when opening via e-mail or file servers, and would validate their signature. A document from the outside world (or one not going through the corporate-mandated signature process) would outright fail to open with an error message and instructions to ask the sender to please sign the document.

(Lots of details to work out there, but if this process could be made simple and mostly automatic, you'd help close a major attack vector that companies are susceptible to today.)

Anyway, it's great hearing your thoughts on how Keybase plans to make money. I've been in the same boat of loving Keybase but being uncertain about where it'll be 5 years from now. We'll keep an eye open for some paid products :)

A large portion of their code is open source: https://github.com/keybase

transfer.fsckObjects=true

There's been talk of making this the default, but it's trivial enough to stick in your .git