> [mailto:[EMAIL PROTECTED] On Behalf Of Dave Crocker > Steve Atkins wrote: > > A lot of the controversy about SSP is based on false > positives - mail > > that was signed when sent but is not signed when received. > > > > I know that various people have been looking at the cases > where that > > can happen, but I don't recall seeing any quantitative results > > presented. If they have been, could someone point me at them? > > I wonder whether this issue might be amenable to resolution > without worrying about empirical statistics. > > In other words: Perhaps one or all SSP settings require > non-breakage along the path. If there is breakage, the > mechanism is effectively disabled. > > In other words: Exactly how bad is it, for a legitimately > signed message to fail the signature check and then be > subjected to the usual vagaries of filter analysis? > > Similarly: Exactly how bad is it for an SSP I-Sign-All > domain to have an unsigned message succeed through filters > and get delivered? (I am tossing this into the mix, from the > exchange Delany and I are having, about dictating delivery behavior.)
I think that the sensible mail filtering product providers (i.e. the ones whose stock grants may end up worth something) would generate their own statistics. So I would expect to see the filters at large ISP note that if an email purports to come from Ebay and its not signed and Ebay has an 'I sign everything' policy then its probability of passing the spam filters is remarkably low (1-2% ?)and the chance that it will be reported as phishing if delivered is very high (compatible with 100% spam). One value of having the policy in place is that Ebay can stop signing or change its signing profile without being clobbered by filters that have learned to expect that Ebay signs everything but do not have an explicit statement to that effect. In other words all the ills that can flow from having policy that is misinterpreted can also flow from not having policy and in many ways these results are worse. I would certainly hope that the filters themselves have the ability to note things like '100% of the messages I attempt to verify turn out to be broken, therefore I conclude that there might be something wrong with incomming email and will tell my operator to check or possibly request a test message from a known good source.' We have a reactive system here. DKIM is intended to change the infrastructure of email and the attackers are attempting to stop it. So statistics are good at the level of telling if something is a 1% effect, 5% effect 20%, 80%, 95%, 99%. But guessing how they will react is just that. _______________________________________________ NOTE WELL: This list operates according to http://mipassoc.org/dkim/ietf-list-rules.html