Email > Spam Filtering > System Spam Filtering

System Spam Filtering

CSLab's mail system can automatically filter spam for you in a number of ways, for your own personal address and for your mailing lists; you can use different settings on different mailing lists and on your own address. This filtering is managed through our self-serve email anti-spam settings pages.

We support a number of options for this filtering:

* No automatic system spam filtering

• No automatic spam filtering will be done, just the standard PureMessage spam tagging and virus removal. You can use personal email filters to do any filtering that you want.

* Moderate (system) spam filtering

• The system will automatically reject email that it feels has a high chance of being spam. Because we reject it, senders will always get notified that their message did not get through.

* Strong (system) spam filtering

• The system will reject or discard email that it feels is spam. Senders will not always get notified that their message did not get through.

* Only accept email sent from DCS computers

• We will only accept email for this address from machines that we consider to be inside the Department of Computer Science (including CDF, DGP, and so on, not just CSLab-managed machines); email from other machines will be rejected. Senders will always get notified that their message did not get through.

* Only accept email sent from University of Toronto computers

• This is as above, but for University of Toronto machines instead of just DCS machines.

It is important to note the difference between machines and people; for example, someone in the department might send you email from a Google Mail account, and that would be considered to be outside the department because GMail's mail machines are not inside the department.

For moderate or strong filtering, you can opt to have email sometimes be delayed in order to get less spam. In addition to sometimes delaying email to your address, it is possible for a sender's email system to dislike this and to fail to deliver the message; if this happens, the sender should be informed about it.

(If you are familiar with email anti-spam technology, this is known as 'greylisting'.)

Messages that are automatically filtered out by the mail system will not be put in your oldmail.

The specific techniques that we use to implement moderate and strong filtering may change over time. We guarantee that moderate spam filtering will only ever use techniques where the sender is informed if their message doesn't get through. With strong filtering, there is no such guarantee, although we will try to have senders informed as much as is practical.

We feel that moderate spam filtering (with or without delaying email) is pretty safe and quite effective. Strong mail filtering is more dangerous since if a message is mis-classified as spam (which does happen from time to time), no one will ever know about it.

Technical details

IMPORTANT NOTE: The exact technical details of what specific filtering is done for each option above is subject to change over time, as we evolve our anti-spam systems in order to make them more effective. While we will try to keep this documentation up to date, we do not promise that it will always be complete, and we will certainly not be notifying people every time it changes. If you opt in to system spam filtering, you are opting in to our best faith implementation of the above ideas, not some specific technical filter.

That being said:

As of July 19th 2011, moderate spam filtering rejects email from machines listed in the CBL or in Spamhaus's lists; in addition, it will sometimes reject email that PureMessage thinks has a virus or that PureMessage gives a spam score of 98 or higher to. Strong spam filtering also discards messages that PureMessage scores as spam (for whatever internal reasons PureMessage decides on); in addition, it can sometimes reject such email instead of accepting and then discarding it. As mentioned above, the 'delay email for a while to avoid (some) spam' option is ordinary greylisting.

Rejection is done during the SMTP conversation in reply to RCPT TO and DATA commands from the sending mail server, and so should always cause error messages to be delivered to the message's author. For technical reasons, PureMessage based filtering can only sometimes reject messages during the SMTP DATA phase; otherwise it is limited to discarding messages after we have accepted them during the SMTP DATA phase.

(Whether PureMessage-based rejection is possible depends on who else the message is addressed to at the SMTP level. Because they happen in reply to a SMTP DATA command, these rejections apply to all recipients and so the system can only pick the weakest level of rejection that all recipients have approved. So if a spam message is sent to someone with strong spam filtering and someone who has not opted for any spam filtering, we have to accept it so that it can be delivered to the second person and then discard the first person's copy later.)

Because spammers often forge the sender information on their messages, we cannot safely generate bounce messages when we discard PureMessage scored messages. The only safe time to generate rejections is during the SMTP conversation, where it is the sending mail server's responsibility to decide what to do.