[Cryptography] phishing, was Encryption opinion

ianG iang at iang.org
Tue Sep 9 22:21:05 EDT 2014


On 9/09/2014 16:50 pm, Jerry Leichter wrote:
> On Sep 8, 2014, at 11:45 PM, John Levine <johnl at iecc.com> wrote:
>>>   "Comcast Wi-Fi serving self-promotional ads via JavaScript injection"
>>>
>>> If carriers and others are permitted to inject anything the framework
>>> for MITM attacks is established and made legal.
>> This battle was lost long, long ago.  ISPs routinely intercept
>> NXDOMAIN DNS results and replace them with A records pointing to a web
>> server with "helpful" paid ads.
> I've always found this one complicated.  If done without the end user's permission, it's an issue.  But there are people who believe it's bad even *with* the end-user's permission - for reasons I find hard to follow.


Just from an opinionated pov:  If it is done through a click-thru
licence, then it's not really done with the end user's permission, only
with their trust that the service won't screw them.  That's because an
end-user is not expected to nor can read the licences in modern Internet
software.  It would be an argument in court as to whether the above
'adjustment' is reasonable or not, but what is not really sustainable is
that "end user's permission" exists unless the end user has been
explicitly asked the question and there is recorded evidence to that effect.

E.g., this would pass the reasonable test:

>> On mobile networks, it's very common
>> to force web traffic through caches, and to re-encode images to
>> smaller versions appropriate for the tiny screens.

> ...  For a simple point to point connection between Alice and Bob, who the "man in the middle" is is clear.
>  When a single page on the Web has contributions from tens to hundreds of servers for ads, tracking, and who knows what - not to mention multiple network and edge service providers - it becomes very vague.

This is a key point.  MITM was introduced to the net from various
sources as a key threat.  The one big threat that would tear everything
apart.  So people responded to it.

Things have changed, we're now 20 years on and we're not talking about
single threats, single packets, single connections indeed single
anythings.  A protocol that considers MITM as its prime threat is naive.

What should we be protecting?  It ain't the MITM.  So how do we find a
focus or a goal that allows us to develop systems that better do
something of help to users?  It seems like the classical threat
modelling ideas have run out of grunt.


>> ...There's a couple of companies with names I forget (no doubt they want
>> me to) who produce specialist equipment to do DPI and replace packets
>> like this.
> http://arstechnica.com/tech-policy/2014/09/meet-the-tech-company-performing-ad-injections-for-big-cable/
> 
>> Most of the web nonsense can be prevented with https, at least if you
>> have an honest browser that doesn't interpret https as "tell the cache
>> at the ISP to fetch the https version".
> Indeed.


Yes.  What do we have to do to get back to the original design?
Everything HTTPS?


iang



More information about the cryptography mailing list