You are viewing a plain text version of this content. The canonical link for it is here.
Posted to users@spamassassin.apache.org by Jon Ribbens <jo...@unequivocal.co.uk> on 2006/12/19 17:09:59 UTC

What to do about False Positives on messages I am sending?

I work at a company with an automated on-line system. This system
sends emails to people. Spam Assassin appears to be triggering very
strongly, and incorrectly, on our messages.

FWIW, no we are not spammers, in fact the emails I'm talking about
aren't even a mailing list. They're emails generated in response to
a (confirmed) registered user performing an action on the system
(each email goes to a single recipient, not bulk).

A couple of examples of the tests being triggered include:

  EXTRA_MPART_TYPE

  This one appears to be penalising people who comply with the RFCs.
  multipart/related *requires* the 'type' parameter that is being
  flagged as 'spammy'.

  TVD_FW_GRAPHIC_NAME_MID

  This one appears to be penalising people who put images in the email
  with sensible names.

  HTML_IMAGE_ONLY_12
  HTML_SHORT_LINK_IMG_2

  These two appear to be penalising people who send short messages.

I have read the AvoidingFpsForSenders page, and I am already doing
most of what it says. I'm not encouraged by the first point:

  "The rules catch spam. If your email isn't spam, you shouldn't be
  matching the rules."

I don't see how you can claim this with a straight face, given the
rule examples I've mentioned above. One of the later bits of advice,
"If you're using HTML emails, include a text part" is precisely what
is triggering your own "spam-detecting" EXTRA_MPART_TYPE rule!

I could work around these problems - I could break the RFC rules to
avoid EXTRA_MPART_TYPE, I could obfuscate the image filenames to avoid
TVD_FW_GRAPHIC_NAME, I could pad the message with invisible junk to
avoid HTML_IMAGE_ONLY etc. But that would be ridiculous - that's what
spammers do! Am I supposed to disguise my non-spam messages as spam in
order to prevent SpamAssassin calling them spam?

Any advice would be gratefully received! On the plus side, I should
point out that we have recently implemented SpamAssassin on our
incoming email and it's cut down the spam on the 'catchall' mailbox
from approximately 3,000 a day to more like 10, so it's being very
helpful in that respect ;-)

Cheers


Jon

Re: What to do about False Positives on messages I am sending?

Posted by Adam Lanier <ad...@krusty.madoff.com>.
On Tue, 2006-12-19 at 16:58 +0000, Jon Ribbens wrote:
> 
> But that's all a bit philosophical and beside the point of my
> question, which is: should I change our emails, and if so, in what
> way - or do SpamAssassin's default settings as provided on
> updates.spamassassin.org need changing?

Perhaps you could post a sample message so we could take a look at the
structure?

Re: What to do about False Positives on messages I am sending?

Posted by Jon Ribbens <jo...@unequivocal.co.uk>.
Sietse van Zanen <si...@wizdom.nu> wrote:
> Do you have your trusted_networks, internal_networks and all_trusted set
> up correctly?
> 
> With these three options you should be able to exclude messages sent
> from your IP address.

Yes, the problem is not that *our* SpamAssassin installation is
flagging our mail as spam, it's that (some of) our *customers'* spam
filters are flagging our mail as spam, and I assume that many of their
spam filters use SpamAssassin. Obviously it's hard to get all of them
to add our mail server IPs to their whitelists ;-)

> BTW, you are sending bulk mail (same mail, many recipients) and bulk
> mail isn't necessarily spam of course.

Actually the mails I was talking about aren't bulk mail, because they
are different mails with different recipients. They are automated, but
not in the sense of "fill in the name and address into a template and
send 1,000 copies" but in the sense of "do some work which can take up
to several hours to perform, then send 1 personal (contains individual
report results) email to 1 person to report completion".

But that's all a bit philosophical and beside the point of my
question, which is: should I change our emails, and if so, in what
way - or do SpamAssassin's default settings as provided on
updates.spamassassin.org need changing?

Cheers


Jon

RE: What to do about False Positives on messages I am sending?

Posted by Sietse van Zanen <si...@wizdom.nu>.
If you look at politicians you will surely see that saying: "you
shouldn't ..." wih a straight face is not that hard at all. :-)

Do you have your trusted_networks, internal_networks and all_trusted set
up correctly?

With these three options you should be able to exclude messages sent
from your IP address.

BTW, you are sending bulk mail (same mail, many recipients) and bulk
mail isn't necessarily spam of course.

Ultimately you could even separate outgoing and incoming mail, by using
multiple MTA's. Then you can use the outgoing MTA without SA, so it
saves you some resources too.

-Sietse 


-----Original Message-----
From: Jon Ribbens [mailto:jon+sa-users@unequivocal.co.uk] 
Sent: Tuesday, December 19, 2006 5:10 PM
To: users@spamassassin.apache.org
Subject: What to do about False Positives on messages I am sending?

I work at a company with an automated on-line system. This system
sends emails to people. Spam Assassin appears to be triggering very
strongly, and incorrectly, on our messages.

FWIW, no we are not spammers, in fact the emails I'm talking about
aren't even a mailing list. They're emails generated in response to
a (confirmed) registered user performing an action on the system
(each email goes to a single recipient, not bulk).

A couple of examples of the tests being triggered include:

  EXTRA_MPART_TYPE

  This one appears to be penalising people who comply with the RFCs.
  multipart/related *requires* the 'type' parameter that is being
  flagged as 'spammy'.

  TVD_FW_GRAPHIC_NAME_MID

  This one appears to be penalising people who put images in the email
  with sensible names.

  HTML_IMAGE_ONLY_12
  HTML_SHORT_LINK_IMG_2

  These two appear to be penalising people who send short messages.

I have read the AvoidingFpsForSenders page, and I am already doing
most of what it says. I'm not encouraged by the first point:

  "The rules catch spam. If your email isn't spam, you shouldn't be
  matching the rules."

I don't see how you can claim this with a straight face, given the
rule examples I've mentioned above. One of the later bits of advice,
"If you're using HTML emails, include a text part" is precisely what
is triggering your own "spam-detecting" EXTRA_MPART_TYPE rule!

I could work around these problems - I could break the RFC rules to
avoid EXTRA_MPART_TYPE, I could obfuscate the image filenames to avoid
TVD_FW_GRAPHIC_NAME, I could pad the message with invisible junk to
avoid HTML_IMAGE_ONLY etc. But that would be ridiculous - that's what
spammers do! Am I supposed to disguise my non-spam messages as spam in
order to prevent SpamAssassin calling them spam?

Any advice would be gratefully received! On the plus side, I should
point out that we have recently implemented SpamAssassin on our
incoming email and it's cut down the spam on the 'catchall' mailbox
from approximately 3,000 a day to more like 10, so it's being very
helpful in that respect ;-)

Cheers


Jon

Re: What to do about False Positives on messages I am sending?

Posted by Theo Van Dinter <fe...@apache.org>.
On Tue, Dec 19, 2006 at 10:46:22AM -0800, John D. Hardin wrote:
> Do they still subtract points from the score? That's the relevant 
> factor.

Yes, they do.  Just sharing that it doesn't involve modifying the message
anymore. :)

-- 
Randomly Selected Tagline:
"Hey, you know what'd cheer you up? You should get yourself a puppy." -Amy 
 "A puppy? Nibbler loved to eat puppies...." -Leela 

Re: What to do about False Positives on messages I am sending?

Posted by "John D. Hardin" <jh...@impsec.org>.
On Tue, 19 Dec 2006, Kelson wrote:

> John D. Hardin wrote:
> > Do they still subtract points from the score? That's the relevant 
> > factor.
> 
> The headers?  No.  Unless you're running a really old SpamAssassin.

No, the fact that the sender has registered with either Habeas or 
Bonded Sender.

--
 John Hardin KA7OHZ                    http://www.impsec.org/~jhardin/
 jhardin@impsec.org    FALaholic #11174     pgpk -a jhardin@impsec.org
 key: 0xB8732E79 -- 2D8C 34F4 6411 F507 136C  AF76 D822 E6E6 B873 2E79
-----------------------------------------------------------------------
  "Bother," said Pooh as he struggled with /etc/sendmail.cf, "it never
  does quite what I want. I wish Christopher Robin was here."
				           -- Peter da Silva in a.s.r
-----------------------------------------------------------------------
 6 days until Christmas


Re: What to do about False Positives on messages I am sending?

Posted by Kelson <ke...@speed.net>.
John D. Hardin wrote:
> Do they still subtract points from the score? That's the relevant 
> factor.

The headers?  No.  Unless you're running a really old SpamAssassin.

-- 
Kelson Vibber
SpeedGate Communications <www.speed.net>

Re: What to do about False Positives on messages I am sending?

Posted by "John D. Hardin" <jh...@impsec.org>.
On Tue, 19 Dec 2006, Theo Van Dinter wrote:

> On Tue, Dec 19, 2006 at 09:59:40AM -0800, John D. Hardin wrote:
> > ...sign up with a service like Habeas or Bonded Sender and put their
> > headers in your messages?
> 
> FWIW, neither of those put headers in the message (Habeas stopped
> doing that years ago).  They're both DNS reputation services.

Argh.

Do they still subtract points from the score? That's the relevant 
factor.

--
 John Hardin KA7OHZ                    http://www.impsec.org/~jhardin/
 jhardin@impsec.org    FALaholic #11174     pgpk -a jhardin@impsec.org
 key: 0xB8732E79 -- 2D8C 34F4 6411 F507 136C  AF76 D822 E6E6 B873 2E79
-----------------------------------------------------------------------
  "Bother," said Pooh as he struggled with /etc/sendmail.cf, "it never
  does quite what I want. I wish Christopher Robin was here."
				           -- Peter da Silva in a.s.r
-----------------------------------------------------------------------
 6 days until Christmas


Re: What to do about False Positives on messages I am sending?

Posted by Theo Van Dinter <fe...@apache.org>.
On Tue, Dec 19, 2006 at 09:59:40AM -0800, John D. Hardin wrote:
> ...sign up with a service like Habeas or Bonded Sender and put their
> headers in your messages?

FWIW, neither of those put headers in the message (Habeas stopped doing that
years ago).  They're both DNS reputation services.

-- 
Randomly Selected Tagline:
Have I ever claimed to be sane?

Re: What to do about False Positives on messages I am sending?

Posted by Theo Van Dinter <fe...@apache.org>.
On Wed, Dec 20, 2006 at 06:44:48PM +0000, Jon Ribbens wrote:
> I did that. The problem that needs fixing is SpamAssassin. It is
> triggering on things that are nothing to do with spam (for example,
> RFC-compliant use of multipart/related). 

Your main issue is that spammers are making their mails look more like ham,
and that sometimes leads to more FPs.  There's already discussions going on
about things like EXTRA_MPART_TYPE and such.  For example:

https://issues.apache.org/SpamAssassin/show_bug.cgi?id=5226
https://issues.apache.org/SpamAssassin/show_bug.cgi?id=5110

etc.

-- 
Randomly Selected Tagline:
"My teacher says I need cupcakes.  Cupcakes to learn."
         - Lisa Simpson, "Simpsons Safari"

Re: What to do about False Positives on messages I am sending?

Posted by Jon Ribbens <jo...@unequivocal.co.uk>.
Noel Jones <no...@gmail.com> wrote:
> So why not find which rules are triggered by your message

I already did - see my original post at the start of this thread.

> Can't be too hard, spammers do it all the time.

That's my point - why should I have to behave like a spammer in order
to avoid getting classed as one?

> If you're not sending spam, why does it look like spam?  Find out and
> fix the problem!

I did that. The problem that needs fixing is SpamAssassin. It is
triggering on things that are nothing to do with spam (for example,
RFC-compliant use of multipart/related). 

Re: What to do about False Positives on messages I am sending?

Posted by Jo <ml...@winfix.it>.
Jon Ribbens wrote:
> Loren Wilton <lw...@earthlink.net> wrote:
>   
>> Taking a look at that and offering my opinions:
>>     
>
> Thanks for taking the time to have a look at it. Apart from inline
> images though, the other points either don't apply to our emails, or
> don't appear to be contributing to the SpamAssassin score.
>
>   
>> In all honesty, I have to ask: does this mail NEED to be html?  Other than 
>> flashy colors and imbedded images what does it buy you that the text 
>> message wouldn't convey?  Unfortunately html, embedded images, align right, 
>> and flashy colors all end up making the thing look like a typical drug spam.
>>
>> I know that flashy colors and imbedded images are important if you are 
>> sending these to CEOs or other people that never learned to read.
>>     
>
> Unfortunately yes, this mail absolutely 100% does need to be HTML.
> A significant proportion of the target audience includes management
> types, sales and marketing types, etc and the presentation is at
> least as important as the content, if not more so.
>
> Personally I don't have anything against HTML emails if they have a
> text equivalent as well, and it's somewhat irritating that it's this
> precise feature that's one of the things SpamAssassin dislikes :-/
>   
It's not so much about liking or disliking. It's about using tell tale 
signs to detecting probable spam.

Jo

Re: What to do about False Positives on messages I am sending?

Posted by Jon Ribbens <jo...@unequivocal.co.uk>.
Loren Wilton <lw...@earthlink.net> wrote:
> Taking a look at that and offering my opinions:

Thanks for taking the time to have a look at it. Apart from inline
images though, the other points either don't apply to our emails, or
don't appear to be contributing to the SpamAssassin score.

> In all honesty, I have to ask: does this mail NEED to be html?  Other than 
> flashy colors and imbedded images what does it buy you that the text 
> message wouldn't convey?  Unfortunately html, embedded images, align right, 
> and flashy colors all end up making the thing look like a typical drug spam.
> 
> I know that flashy colors and imbedded images are important if you are 
> sending these to CEOs or other people that never learned to read.

Unfortunately yes, this mail absolutely 100% does need to be HTML.
A significant proportion of the target audience includes management
types, sales and marketing types, etc and the presentation is at
least as important as the content, if not more so.

Personally I don't have anything against HTML emails if they have a
text equivalent as well, and it's somewhat irritating that it's this
precise feature that's one of the things SpamAssassin dislikes :-/

Re: What to do about False Positives on messages I am sending?

Posted by Loren Wilton <lw...@earthlink.net>.
> I have attached a sample message to this email. Note, it's just an
> example. This message does not trigger at the 5.0 level, but I know
> messages like this are being blocked by some of our customers. It does
> get a higher score than I would like it to (i.e. 0.0 ;-) ), and
> certainly the rules its triggering make little sense to me.

Taking a look at that and offering my opinions:

1.    Avoid text-align: right.  Common spammer trick used to obfuscate drug 
spams.
2.    Avoid excessively long lines in the HTML.  Typical sign of spammers 
that can't quite figure out how to format a message.
3.    Avoid excessive whitespace on the front of HTML lines.  A sign of 
certain forms of table-layout phish mails.  (You mail didn't have this, I'm 
just pointing it out.)
4.    Avoid inline images if possible.
5.    Avoid downloaded images even more.

In all honesty, I have to ask: does this mail NEED to be html?  Other than 
flashy colors and imbedded images what does it buy you that the text message 
wouldn't convey?  Unfortunately html, embedded images, align right, and 
flashy colors all end up making the thing look like a typical drug spam.

I know that flashy colors and imbedded images are important if you are 
sending these to CEOs or other people that never learned to read.  But if 
you are sending this to the sysop to tell him how well his web site works, 
wouldn't it be just as useful to simply send the report in ascii?  That 
would avoid virtually all of the potential spam signs.

        Loren


Re: What to do about False Positives on messages I am sending?

Posted by Jon Ribbens <jo...@unequivocal.co.uk>.
Adam Lanier <ad...@krusty.madoff.com> wrote:
> That's why I asked to see a sample message.  We could probably give some
> pointers on what is triggering SA.

I have attached a sample message to this email. Note, it's just an
example. This message does not trigger at the 5.0 level, but I know
messages like this are being blocked by some of our customers. It does
get a higher score than I would like it to (i.e. 0.0 ;-) ), and
certainly the rules its triggering make little sense to me.

Re: What to do about False Positives on messages I am sending?

Posted by Adam Lanier <ad...@krusty.madoff.com>.
On Wed, 2006-12-20 at 11:38 -0600, Noel Jones wrote:
> On 12/20/06, Jon Ribbens <jo...@unequivocal.co.uk> wrote:
> > "John D. Hardin" <jh...@impsec.org> wrote:
> > > ...sign up with a service like Habeas or Bonded Sender and put their
> > > headers in your messages?
> >
> > I suppose we could do. Does anyone know how much that costs?
> >
> > It still seems wrong to me though that SpamAssassin is penalising mail
> > that doesn't look like spam, and encouraging people to make their ham
> > look like spam.
> >
> 
> So why not find which rules are triggered by your message and just
> rewrite your message so it doesn't trigger those rules?  Can't be too
> hard, spammers do it all the time.
> 
> I find very few legit messages - even legit marketing mail - get
> tagged as spam.
> 
> If you're not sending spam, why does it look like spam?  Find out and
> fix the problem!
> 

That's why I asked to see a sample message.  We could probably give some
pointers on what is triggering SA.

Re: What to do about False Positives on messages I am sending?

Posted by Noel Jones <no...@gmail.com>.
On 12/20/06, Jon Ribbens <jo...@unequivocal.co.uk> wrote:
> "John D. Hardin" <jh...@impsec.org> wrote:
> > ...sign up with a service like Habeas or Bonded Sender and put their
> > headers in your messages?
>
> I suppose we could do. Does anyone know how much that costs?
>
> It still seems wrong to me though that SpamAssassin is penalising mail
> that doesn't look like spam, and encouraging people to make their ham
> look like spam.
>

So why not find which rules are triggered by your message and just
rewrite your message so it doesn't trigger those rules?  Can't be too
hard, spammers do it all the time.

I find very few legit messages - even legit marketing mail - get
tagged as spam.

If you're not sending spam, why does it look like spam?  Find out and
fix the problem!

-- 
Noel Jones

Re: What to do about False Positives on messages I am sending?

Posted by Jon Ribbens <jo...@unequivocal.co.uk>.
"John D. Hardin" <jh...@impsec.org> wrote:
> ...sign up with a service like Habeas or Bonded Sender and put their
> headers in your messages?

I suppose we could do. Does anyone know how much that costs?

It still seems wrong to me though that SpamAssassin is penalising mail
that doesn't look like spam, and encouraging people to make their ham
look like spam.

Re: What to do about False Positives on messages I am sending?

Posted by "John D. Hardin" <jh...@impsec.org>.
On Tue, 19 Dec 2006, Jon Ribbens wrote:

> I work at a company with an automated on-line system. This system
> sends emails to people. Spam Assassin appears to be triggering
> very strongly, and incorrectly, on our messages.

> Any advice would be gratefully received!

...sign up with a service like Habeas or Bonded Sender and put their
headers in your messages?

Just an idea.

--
 John Hardin KA7OHZ                    http://www.impsec.org/~jhardin/
 jhardin@impsec.org    FALaholic #11174     pgpk -a jhardin@impsec.org
 key: 0xB8732E79 -- 2D8C 34F4 6411 F507 136C  AF76 D822 E6E6 B873 2E79
-----------------------------------------------------------------------
  "Bother," said Pooh as he struggled with /etc/sendmail.cf, "it never
  does quite what I want. I wish Christopher Robin was here."
				           -- Peter da Silva in a.s.r
-----------------------------------------------------------------------
 6 days until Christmas