Bayesian Poisoning

If you are like me, you've been getting a lot of phoney SPAM in your inbox (or SPAM folder). This SPAM isn't really SPAM, it's not selling anything, it is just a bunch of gibberish, or maybe excerpts from famous literature or movies. This line is a real example, "Apparently, the metallic core unit does indeed flow slightly better than the ceramic unit".

Well, the purpose if this SPAM is an attempt at Bayesian Poisoning. I read about this recently at the Virus Bulletin web site, in the article Does Bayesian poisoning exist? And the article goes on to say:

"The evidence suggests that Bayesian poisoning is real, but either impractical or defeatable."

Bayesian filters read your e-mail and try to determine if it is indeed SPAM just by the content, using a statistical content filter. They assign a point score to the e-mail, maybe on a scale from 1-5. And depending on how you have your filter set (mine's at 4), determines whether you will get the e-mail in your inbox or not. Bayesian poisoning is an attempt to ruin these statistical filters by adding words that are not usually in SPAM messages to the filter, skewing the stastical grade of the e-mail, allowing it to get through.

After reading about these attempts, I wondered if I should tell our e-mail users not to forward these types of e-mails to the SPAM filter, for fear of weakening it. But then I read the article and it mostly put my mind at ease. I will continue to tell our Users, if they get this literary SPAM, go ahead and add the mail to your SPAM filter.