Re: move to second stage, Re: Principles of Spam-abatement

Ed,

Thank you for the wealth of information. I forwarded this message to the 
SMTP-VERIFY subgroup of the ASRG where the current discussion of the 
"web of trust" is taking place so we can evaluate the information (the 
archive can be found at 
http://news.gmane.org/gmane.ietf.asrg.smtpverify/). If you or anyone 
else wants to join that list, let me know off-list (its a closed list 
with about 30 members, separate from the main ASRG list).
I will get back to you once we digested your paper.

Yakov

Ed Gerck wrote:
> Yakov Shafranovich wrote:
>
>
> > Go ahead - I am looking for any kind of solutions that the IETF can take
> > on in order to reduce the problem. Many solutions have been revolving
> > around trust - but in the world where a computer can be easily hijacked,
> > trust becomes harder to maintain.
>
> Trust is the problem [1]. What you mention below is a valid way to
> induce trust, namely by relying on trusted introducers (for trusted
> *and* distrusted MTAs). The question of qualifying the trusted 
> introducers themselves is also qualitatively handled in the model you
> summarize. One thing that is missing is what I call the trusted 
> witnesses, which are also necessary to induce trust [2].
> Trusted introducers and trusted witnesses allow you to build two 
> open-ended trust chains for every action, the witness chain providing 
> the assurances ("how did we get here?") that led to action (including
> the action itself) while the introducer chain ("where do we go from 
> here?") provides the assurances both for a continuation of that action
> and for other actions that may need assurances stemming from it.
>
>
> > One example of what the ASRG has been looking at is a distributed web of
> > reputation. Each MTAs or domain can publish a list of MTAs that it
> > knows, including basic statistics on how long the MTA has been sending
> > mail, average volume, etc. In addition to that basic information, you
> > can also publish additional information such as "I think this is a
> > spammer because SpamAssasin detects 99% of all email from that MTA as
> > spam", etc. The basic statistical information can be used to detect
> > zombies and the extended information can be used to allow like-thinking
> > domains to make joint decisions. The question of how much difference
> > this would make is up for debate, and there are questions of how a new
> > MTA can be introduced into the system, "rule of the mob", etc.
>
> Seeing this as a web of trust would seem to clarify the issues you mention 
> and point out what's missing. Relying on reputation alone (and not also on 
> current behavior, etc.) can lead to race conditions, bait-and-switch, 
> spoofing, laggard reactions, and a host of other threats that are easily 
> exploitable. BTW, if reputation alone would be a solution, eBay customers
> would not have so many problems with online auctions -- the Federal Trade 
> Commission says it receives more complaints about Internet auction fraud 
> than any other online scam. Internet auctions accounted for 48 percent of 
> all Internet fraud complaints filed with the commission in 2003 (1/26/04 
> article by Brian Krebs on eBay Feedback Forgers).
> Comments?
>
> Cheers,
> Ed Gerck
>
> [1] Understanding human trust is exactly what brought me to that great
> IT question in 1997: how can I trust a set of bytes? My answer provided 
> a framework that has been useful in the field of information security.
> The answer also provides a framework for understanding human trust 
> (as expected fulfillment of behavior) and bridging trust between humans 
> and machines (as qualified information based on factors independent of
> that information). The original reference is 
> http://nma.com/mcg-mirror/trustdef.htm
> -- please google for "gerck trust" to find applications and comments
> by others.
>
> [2] An example is described in
> http://nma.com/papers/e2e-security.htm#TR
>
>   "...under the principle that every action needs both a trusted 
>   introducer and a trusted witness. We call this principle the 
>   Trust Induction Principle."