Spamanade

So I've been spending some time running batches of spams & hams through spamassassin, and whipped up some decent scripts to automate the whole remote sa-learn/archiving deal. Works like a charm so far.

I have to run spamassassin at a fairly... modest... trigger level, but it still does a decent job. What I've found to be amusing are some of the subjects of the spams that have actually gotten through. If you haven't followed this, most of the anti-spam tools out there are moving to teachable metric system for detecting spam. It has a basic set of weighted rules for trying to guess what is/isn't spam, but it often needs some fine tuning.

IE, you might be getting spam ezines for marketing your website, but someone else might actually be requesting iframe-laden ezines on how to market their website, but a general spam filter just knows what spam might look like, without the conditionals (ie, that it was subscribed to). So you can generally teach these filters by running through incorrectly identified spams (hams) and it looks at various characteristics of the message (headers, to/from, body, etc) and gets a better understanding of the type of mail you actually get.

Spam has gotten so bad that antispam tools are becoming increasingly commonplace both at the server level and the client. Most email clients either have or are planning on incorporating it. In all honesty though, lots of these never get fine tuned, and even then they're fallible... they just aren't very fallible as they've been building up their metrics for a long time.

So what's an enterprising spammer to do? Create a program to generate extremely weird subjects, often at random, often with odd characters... anything to throw off the spam filters metrics. They basically just sit there and run a million different variations of a message through the spam filters until they see what gets through on a default install. If something gets through... you've got yourself a golden message that has much better chance of getting through the filters. I know this is the idea, although I haven't really looked into what tools they're using to do it. I'd actually wonder if they couldn't use a form of evolutionary algorithm...

...but anyways. The really odd combinations of words that get produced can really help when you are having to visually fish out the spams that have gotten past the future. After seeing enough of them, your eyes start to kinda glaze over. It can be a real chore. So when some enjoyment does present itself, I'm all over it... in this case, I just started saving some of the really odd subject lines I've gotten through these before they get passed into the filter.

Before I'd save it, it had to pass two criteria: it had to amuse me, it had to get past spamassassin. There were some with some amusing message bodies, but those didn't count.

So far: