Re: [Asrg] RMX proposals and Nash Equilibrium

Daniel Feenberg <feenberg@nber.org> wrote:
> Given that Hotmail participated, would I set my mailer to check MAIL FROM
> RR records? If sendmail supported it as a feature, I probably would. It
> would cut 50% of my incoming spam, with very few type two errors. So in
> the short run I have a substantial positive incentive to do the checking.
> But remember that most mail won't come from checkable addresses. I can't
> start rejecting that mail without justifiable complaints from my users.

  That's not the point.

  The usual idea is that any new anti-spam method allows MTA
administrators to catch & delete spam.  This idea is wrong.

  Anti-spam methods allow MTA administrators to make better decisions
with less work, than their curent methods.  If one of the outcomes of
those decisions is to drop mail, then it's a local MTA issue, and
totally unrelated to the anti-spam method.

  An alternative policy approach for sites implementing RMX lookups
would be to rate-limit non-RMX connections to some small amount of
bandwidth.  Mail from non-RMX sites would still get through, but would
be punished relative to RMX aware sites.

> Would I have much incentive to register my server names? To start with,
> very little, as I am not aware of any spammers using my return address,

  You're lucky.  Nothing more.

> But there are millions of possible names. Spammers could use all of them -
> they already have them, after all. At that point the incentive for sites
> to check incoming mail is greatly reduced. If there are any disadvantages
> (and there are some related to roving users and forwarded mail) then a
> dramatic reduction in the advantage might sink the proposal at
> non-participating sites. It won't catch 50% of the mail anymore , but only
> a tiny fraction.

  What you're missing is that email is not evenly distributed across
domains.  So if even 20% of mail to an ISP comes from a few sites
implementing RMX, that 20% will contain little to no spam.

  In addition, RMX still allows nearly zero-cost "grouping" of
messages into categories:  accountable originating MTA's, and
non-accountable ones.  At that point, it becomes feasible to drop
nearly all content filtering for accountable originators, because you
can trust (at some level) that any spam from them will be
non-deniable, and that they will probably deal with the issue.

> Therefore I believe that the eventual equilibrium is that many domains may
> register allowed SMTP sources for their outgoing mail, and some will
> reject incoming mail which does not match one of the registered domains,
> but that essentially no sites will reject mail from non-participating
> domains and spammers will not use registered domains in MAIL FROM:
> addresses. Net improvement:  little or none.

  Nonsense.  Again, you're stuck on the idea that the MTA choices are
to drop the mail, or to accept it blindly.  There are more options.

  Non-RMX sites can be put through a content filter which has a high
false positive rate, but a low false negative one.  Any bounce/error
message from the content filter can say "Because you didn't do
anything which allowed us to trivially verify you, we're doing the
work, and we're doing it in a way to make us happy.  If it means
dumping your messages, well, it sucks to be you."

> Compare this to connection IP address based blacklists of open relays. In
> that case each participating site benefits from refusing mail from open
> relays.

  Again with the accept/refuse...

> As relays are closed the remaining relays share the traffic among a
> smaller group. The increased traffic increases the incentive for the
> remaining relays to close up.

  True, but irrelevant.  Little spam comes from open relays.  A lot
comes from knowingly complicit senders, from "owned" machines, or from
throw-away accounts.

  Alan DeKok.
_______________________________________________
Asrg mailing list
Asrg@ietf.org
https://www1.ietf.org/mailman/listinfo/asrg