[Spambayes] spambayes...

[Neil Schemenauer]

wsy at merl.com wrote:
> The problem is this: the "from" address on spam is almost always forged.
> 
> That doesn't make the task impossible, just harder.  You need to execute
> the bounce at SMTP transfer time, not at MDA execution time (which is 
> where filters usually go).

Right.  Bouncing the message after the SMTP transaction does more harm
than good.

> Basically, one needs to write an SMTP front end that does filtering and
> 400's the message during the SMTP protocol whenever it thinks it's spam.

I have something for qmail:

    http://arctrix.com/nas/qmail/spambayes/

Greg Wand and I also have something for Exim (using elspy).  I've placed
a sanitized version here:

    http://arctrix.com/nas/exim/local_scan.py

It should be generally usable with a little editing.  Note that both of
these implementations use my "spam delay" idea.  I.e., if the message
looks like spam, return a temporary 4xx error for a certain amount of
time.  The Exim version is more lenient because it is being used in a
corporate setting.

I've toyed with the idea of writing an SMTP proxy that does filtering
using spambayes and then passes the message to a real MTA.  Would people
find this more useful than a MTA specific solution?

  Neil