Spam Filter De-Training

August 10th, 2006 by Potato

I noticed that over the last few months I’ve gotten a lot more spam coming through to all my addresses. I was kind of annoyed since I got the same messages a few times after I already plunked one version in my quarantine, so the filters weren’t learning too quickly… but what was stranger that a lot of it was getting by the filters because it wasn’t even trying to sell anything. Some of them didn’t even have links. I thought it was really strange and figured that there was a malfunctioning spambot out there somewhere, or that someone was just sociopathic and liked to clog up the tubes with spam just for the love of making strangers angry. I’ve seen a few other neat theories on this out there, such as a semi-sentient artificial intelligence attempting to communicate with us by using spam — the most common form of communication on the internet (even before Chuck Norris jokes).

Just tonight I saw an even better theory: that the weird snippets of somewhat-normal looking text with no links is being sent around to de-train spam filters. They are messages that are quite similar to normal messages, so they slip through the filters as they stand. If you use those messages to train your net, then the odds are good you’ll increase your false positive rate, and possibly become fed up with your filters and start lowering your threshold, allowing more marketing spam through…

Spammers can be so evil.

Comments are closed.