On a lark, I checked my briefly-used GMail account to see what it had accrued in my absence. Of those unfiltered, what I noticed were a large number of spams that followed this general formula:
did you hear abot that little device for
decoding all the channels that Myles got
last week he says it works real good and he is
watching all thes ppv movies and sporting events
for nothin…LOL…i thought ya right but it actually
check it out if you want at this place [link]
but if you dont want thats fine as
you can stop by and tell us to not tell you anymore [link]
To be fair to Google, I had deliberately seeded that address to fish for 419ers, the [baiting of whom](http://419eater.com/) was a hobby I briefly considered engaging in. Several ‘dead’ address books were signed as a wealthy Australian surgeon, with a fat portfolio ready for his retirement.
###Myles And His TV For Nothin
So… I never knew anyone named Myles, and neither did my Australian alter ego. It’s an interesting approach to spam, because not only does it appear to be in compliance with the CANSPAM regulations, by including unsubscription verbiage, but it’s also actively attempting to defeat statistical analysis without including an obvious body of ‘innocent words’.
Phrases like “last week”, “real good”, and “stop by” probably give this message the green light under current filtering schemes. And frankly, I’m not sure if I want messages that look like [block of informal text + link] to be filtered, because I get a lot of those.
###The Final Spam Filter
What does the Ultimate Iron-Clad Spam Filter look like? I think what it does is crawl the email for links, _visit those sites_, and then analyse _them_ for spamminess. Probably even keep a certral index of ’sites that spams link to’, the inclusion of a link to which is the ultimate damnation for a message.
As I’ve [written previously](http://uwmike.com/archive/antispam-measures/), blocking spams by source-IP is not an acceptable solution. However, I think blocking them by destination may just be the silver bullet. Yes, a spammer can buy up 200 domains to rotate through his spams, but when the first couple folks report the mail as junk, they’ll be quickly flagged. And that assumes that the actual text of the site is clean enough to not get flagged by the content-based filter.
Paul Graham suggested that the [spam of the future](http://paulgraham.com/wfks.html) would look like “Hey, check this out: [link]“, but it’s interesting to see that they’re actually beyond that; they’re using harmless language in the actual body of the message in order to couteract the inherent spammyness of including a link.
###And Over on Stage Left
There may be radically other solutions in the pipe, however, such as the vicious [Project Honeypot](http://www.projecthoneypot.org/), which I’m a participant in, having donated a subdomain and installed a honeypot here. Spammers must have nightmares about their harvest bots going awry and scooping up those innocent looking honeypot addresses.