> Hi,

> Stephen Farrell <[email protected]>:

> #Not quite sure, but I think we might get some benefit at the
> #moment from considering how specific fields in real protocols
> #undermine privacy (e.g. as Christian's draft does with the
> #Received header fields in mail messages) even if/when TLS or
> #other existing security mechanisms are properly used.

> My concern is that many traffic analytic approaches tend to be
> exceedingly robust to "protocol improvements." Protocol tweaks
> may accomplish little when it comes to practically improving
> privacy if the underlying protocol's architecture and operational
> practice goes unchanged.

> For example, when it comes to email, shouldn't section 6 of
> http://huitema.net/papers/draft-huitema-perpass-analthreat-00.txt
> basically say, "if you want to avoid traffic analytic approaches
> in the case of email, deploy and use Mixmaster anonymous remailers"?
> ( https://en.wikipedia.org/wiki/Anonymous_remailers#Untraceable_remailers )

And good luck with that, at least on any kind of scale.

But your underlying point is very well taken: The section on email in this
draft focuses on irrelevancies and fails to take note of the real issues.

I hate to sound like a broken record, but folks really need to have some
familiarity with present-day email as it is actually deployed before making
these sorts of asssessments.

Again, present day email usage is increasingly concentrated to a fairly small
number of large ISPs and MSPs. (Small ISPs and enterprise setups are shifting
to using cloud services, and while the Snowden revelations may have slowed this
trend, they haven't stopped it.)

In regards to traffic analysis, this is in some ways a good thing. If the
connections from user clients to the ISP/MSP servers are secured at the
transport layer - and I have demonstrated that a lot of them are - then we
gain a lot by securing the streams between the large providers at the
transport level.

But the elephant in the corner is logging. Service providers maintain very
extensive logs of email traffic, if for no other reason than as a support
tool. These logs provide every possible detail needed for traffic analysis.

Of course one of the earliest Snowden revelations was that the NSA is
collecting these logs from US providers on a massive scale. And hopefully
everyone is aware of Smith v. Maryland, which essentialls says that metadata
is not constitutionally protected.

But before Eupopeans and others get all smug about this, speaking as someone
who has seen quite a few RFPs for mail systems, the only substantive difference
I see between the US and elsewhere is the US approaches this in a less
organized and systematic way and generally has fewer auditing and data
protection requirements. The data is still being collected, and most likely
shareed.

And as for practical and deployable measures that can be undertaken to address
this, I'm at something of a loss to suggest anything. Shifting back to a more
decentralized model sounds nice, but seems a bit outside the purview of a
standards process to try and make that happen.

And even if it a completely decentralized model was practical, in a
peer-to-peer world the metadata that would accrue from watching the connections
themselves would be a fair substitute.

As for mixed models, look at what happened to Lavabit.

> And if we *are* talking about that sort of approach, then I think
> inevitably we also need to talk about how we simultaneously manage to
> allow *wanted* private traffic while simultaneously preventing or
> managing *unwanted traffic* (e.g., spam).

Yep. It's a daunting problem. And it is far from the only one.

> An awful lot of current anti-spam technology depends upon either
> reputation (which is obviously not present in the case of
> anonymous/non-attributable traffic), or content analysis (which
> is also obviously problematic, at least if we presume use of
> end-to-end encryption (at least until the content is decrypted
> on the end-user's device)).

You basically have to push the content checks to the client. This has
not proven to be a terrific solution in practice.

> I also think that if you're serious about email privacy, you
> really can't keep the discussion just at the level of sanitizing
> headers. You need to get into the format of the content that's
> allowed as well. For example, it's well known that non-plain
> text email content (e.g., HTML-formatted email) is potentially a
> serious threat to privacy due to potential use of things like
> tracking gifs included in HTML-formatted email.

I think we can do a lot to make it harder to snoop on email content, although
ironically what we're likely to be able to accomplish under the "prism-proof"
rubric is unlikely to much of anything about the data collection the actual
Prism program performs.

But traffic analysis... unless the fact that those logs are likely to only be
accessible to state entities offers some consolation, I don't think there's
going to be much happiness here.

                                Ned
_______________________________________________
perpass mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/perpass

Reply via email to