Search Mailing List Archives


Limit search to: Subject & Body Subject Author
Sort by: Reverse Sort
Limit to: All This Week Last Week This Month Last Month
Select Date Range     through    

[liberationtech] What I've learned from Cryptocat

Tom Ritter tom at ritter.vg
Tue Aug 7 06:30:52 PDT 2012


I agree with a lot of the points being raised, including all of
Moxie's (especially about Google v Riseup) but also Eleanor's
regarding niche products and irrelevancy.

In particular I want to expand on this bit:

On 6 August 2012 22:29, Moxie Marlinspike <moxie at thoughtcrime.org> wrote:
> Let's stop using the word "plaintext," because my understanding is that
> none of the chat services we're speaking of transmit data in the clear.
>  As I see it, there are currently three possible vectors for attack with
> "existing" web-based chat services:
>
> 1) SSL interception.
> 2) Server compromise.
> 3) Server operator.
>
> The technology in CryptoCat v1 does not address any of these three
> vectors, and all of them remain possible.  My position is that it's
> actually more susceptible to attack via #1 and #2 than existing
> web-based chat solutions.

I agree with that position wholeheartedly.

Still, possible does not equate to easy.  Cryptocat is the Jackie
Robinson of Web Crypto Services[0].  And not to fault Nadim, as this
is a volunteer effort, but there is more it can and should do to make
it harder for #1 - just as Google and Facebook have done.

 - Cryptocat should use DNSSEC - even if validating resolvers are not
deployed, it's another piece.  Maybe down the road when a binary
plugin is developed, it can validate the DNSSEC chain.[1]
 - It should Pin certificates in Chrome.  As soon as the header is
supported in any other browser it should use it.  Same with TACK.
 - It should assert the SSL certificate with both DANE and
DNSSEC-Stapled Certificates
 - It should (it does for the record, just saying) use Strict Transport Security
 - It should (it now does as of Sunday) deploy Content Security Policy
 - It should do all the other security techniques recommended:
x-frame-options, X-Content-Type-Options, etc
 - Where it is possible, plugins should assert the validity of the Code and Keys
 - Controversial: It should use per-request mutated javascript
obfuscation to make it more difficult to inline-middle the application
in realtime[2]
 - It could experiment with browser enhancements to provide signed
javascript files and so on

All of those are not too difficult for an individual to try, and it
makes SSL Interception harder.  It's not any less possible, but it
raises the bar.  If you think these techniques aren't effective - I
challenge you to do a live MITM of the gmail interface and have it
still function seemlessly.  It just flat-out won't work in Chrome of
course, but even in Firefox - it is not trivial.[4]  That's what I
mean by being Jackie Robinson - you just have to be 'better'.  Above
and beyond.

Trying to fix #2 is way, way harder because it's just impractical to
tell someone "Oh, compile everything from source, run a perfectly
secure Linux box with PaX and grsecurity, etc etc"  Ideally, that's
what we'd have.  I suppose the important part is to acknowledge the
risk of server compromise, and keep the bars approximately equal.  If
all the above measures were employed, but the server left the way it
was - then rooting the box is absolutely easier.  Google and Facebook
have a huge advantage here.


[0] If you don't get the reference, Jackie Robinson broke the
segegration barrier in US baseball - he was spit, cursed, threatened
with murder, had pitches thrown at his head - and through it all, he
just played the game steadfastly. There's an urban legend (I'm unsure
on it's truth) that his contract stated he couldn't complain, even
when fans spit on him.
[1] If your arguement is 'I don't trust DNSSEC' my response is 'Me
neither, but I believe in beaurcracy and turf wars and that it'd be
more difficult for a government to subvert two PKIs than one.'
[2] If your arguement against it is "but then I can't audit it" my
argument is "you don't audit it now." If your argument is still "I
can't trust the code delivered" my response is "well, since you
obviously are doing local SSL interception to read the server
responses, hash the mutated javascript serverside, and send a PGP
signature along with it so you can do the same and trust it that way".
[3]
[3] Heck, maybe that's a way to upgrade web services to thick client
services - have an optional thick client someone can run that sits as
a proxy running verification.  Same web interface works with or
without the tool, but the tool provides run-time verification.
[4] Javascript obfuscation makes it difficult to rewrite the code; if
you add code - CSP prevents you from exfiltrating easily; and if you
exfiltrate to another user in the chat the chatters *should see* this
other user in the chat they don't trust.


> I believe your position is that it improves
> on vector #3 by virtue of being not-Facebook.  (I'm curious how you
> measure #3 in comparison to GChat.)
>
> If we postulate that CryptoCat does improve vector #3 by virtue of being
> not-Facebook, it isn't a result of the technology, but simply that we've
> agreed Nadim has a better monitoring/interception track record than
> Facebook.  If that's something you think is valuable, it actually seems
> like it'd potentially be better served by having someone like the EFF or
> Riseup host a web-based and SSL-protected chat service, without brining
> any additional cryptography confusion into the mix.  A trust project,
> not a cryptography project.

Yes, yes, yes.  There is a *tremendous* amount of implicit and
unmentioned TRUST in the person operating the service or relying on
the software.  That's why anyone would use RedPhone, TextSecure or
WhisperCore back when it was closed source.  Because people *trusted*
Moxie.

If the EFF hosted an e-mail solution I'd be throwing money at them to
let me sign up.  Because Google is huge and diverse - they are
obligated to respond to legal threats in most of the countries of the
world.  Because Google is huge and opaque - I have little faith in
their ability to notify me/etc in the event of a LE request.  But the
EFF has both juristiction on their side (arguably they'd have more if
they were based in, say, Scandenavia) and they have trust.  I trust
that the EFF will fight harder for me than Google.  And there's
intimidation factor - a LE agency ought to know if they come the the
EFF with a request they need to back it up with a supoena or warrant.

I absolutely think of riseup as a trust project.  I would like to see
many more of them.  Nick Merril's Calyx Institute I think of as a
Trust Project.  He's trying very hard to remove Trust, and make it a
cryptography project - but he literally has to build the
infrastructure for that because it doesn't exist.  (Which is one of
the reasons you can't actually buy any services from them yet.)   It
will be *very* interesting to see how Phil Z's and Jon Callas' Silent
Circle positions itself.  A trust project?  They're aiming to remove
trust also; but to what extent can they?

Trying to improve upon the trust factor is extraordinarily difficult.
I think, in the short term, it relies on linking up with a person or
organization people already trust - and therefore somehow convincing
them to trust you. And in the long term - devoting your life to being
a trustworthy individual.  Not something we can solve with
cryptography - even a thick client.

Anyway, I realize I haven't addressed the issue of 'Should cryptocat
move to this model or that, shut itself down, add warnings, push
forward with users, etc'.  But I wanted to raise the point that it can
do more, today, to make users safers - and if _any_ webapp in this
sphere wants to push the envelope, it should probably do those things
first.*  And since people are already using it, these are options to
improve security besides just shutting it down.

-tom

* None of this is meant to be a slight at Nadim, who has certainly
earned my respect for both his effort and his results so far.



More information about the liberationtech mailing list