[tor-talk] Comments?

[Paul Syverson]

On Mon, Aug 07, 2017 at 09:32:20AM +0200, Jon Tullett wrote:
> On 4 August 2017 at 02:05, Paul Syverson <paul.syverson at nrl.navy.mil> wrote:
> > On Thu, Aug 03, 2017 at 04:38:49PM -0700, Jacki M wrote:
> >> Comments on Paul Syverson Proposed attack?
> >> Paul Syverson - Oft Target: Tor adversary models that don't miss the mark
> >> https://www.youtube.com/watch?v=dGXncihWzfw
> >>
> > More seriously. The point of this work is not to propose attacks per se
> > but to observe that a Tor adversary intending to target individuals or
> > specific groups might be much more effective against those targets
> 
> Enjoyed the video - thanks Paul. It cut off before Q&A...were there
> any particularly good questions from the audience?

Ermm probably? I just don't recall completely now weeks later, though
I remember we did fill up the remaining time. Let's see, I had an
exchange with ermm maybe Prateek Mittal, where in response to his
question I emphasized that we weren't suggesting ignoring the
prospects of a hoovering adversary. In fact hoovering might be the
source of some of the inputs an adversary might use to decide to
target an individual, group, or destination. But, it would then be
more likely for the reasons I presented, for an adversary to follow a
targeting strategy even if it had hoovering capabilities already set
up.

Another point that was raised to me in a side comment after the Q&A
was that what I said about evidence for onionsite fingerprinting from
the Oakland poster by Juarez et al.could have been interpreted
differently than I intended.  They showed in that work that
identifying a circuit as an onionsite connection (vs. not an onionsite
connection) via fingerprinting from the middle relay on the client
side was 99.98% effective. They didn't address fingerprinting specific
sites therein. I stand by our claim that given the smallness of the
space and other factors that such fingerprinting is likely for current
Tor and onionspace to be very effective. (Once we make all my
grand-vision changes to onionspace that I'll talk about in my ESORICS
keynote paper, that could well change. But that's c. 5-10 years away
at this point.) Which leads us to...

> 
> Curious to know - at a practical level, have you actually tried any of
> it in practice, or had any contact with anyone who has? I mean, do the
> results in practice mirror the predictions in terms of distribution
> and probability ito deanonymising potential targets?

No we haven't yet tried this in practice. As usual, it is hard to set
up an experiment like that in a way that adequately protects user
safety and privacy. There is work in progress that has gone through
review by the Tor Research Safety Board and will use PrivCount to
gather information about the amount of activity on a particular known
and public onionsite.  But it's still very much in the works.

HTH,
Paul