Anti-spam Tool Kit [Paperback]

This book is the first to follow in the genre defined by "Anti-Hacker Tool Kit." While AHTK examined tools used by intruders and defenders, ASTK focuses on tools and techniques to counter unwanted email. AHTK has slightly more coverage of Windows than UNIX applications; ASTK spends more time on Windows, especially with client configurations. As a desktop FreeBSD user, I still found plenty of helpful information.

Since I'm not directly responsible for enterprise anti-spam defenses, ASTK's comparison of the many block lists was particularly enlightening. Prior to reading ch 5 I knew of many lists but not their particular capabilities. I found the coverage of SpamAssassin in chs 6-8 fairly thorough. Ch 14's overview of email and headers very helpful. I had not heard of server-based greylisting until reading ch 15.

I found a few of the authors' comments odd. On p. 250 they claim "most organizations do not run UNIX-based email solutions." I find this difficult to believe, but no proof for the statement is given. Figure 2.2 in ch 2 shows a mail sender using POP/POP-SSL/IMAP/IMAP-SSL to send email. At least using Mozilla and Firefox on UNIX, I connect to my ISP's SMTP server to send email and then retrieve it with POP or IMAP. In ch 3 the authors advocate IPv6 because it makes "a spammer's identity known and verified even if he or she doesn't want it to be." IPv6 should not change many, or any, spammer tactics. Spammers aren't forging TCP sessions with IPv4 now; they forge information (headers, etc.) transferred within application layer data. I believe some of these weird comments could have been addressed by more diligent technical editing.

Overall, I found ASTK enlightening and practical. The authors give enough details on various server- and client-side tools to make implementation (at least on a trial basis) possible. I look forward to other tool-related books in the "Anti-Hacker" series.

Bayesian techniques are heavily covered here, along with mention of several proprietary versions. The other main technique of note is Realtime Blacklists (RBLs).

But note that the above, and indeed other methods mentioned, suffer from the drawback of heavy manual intervention. For example, to periodically retrain a Bayesian on new sample sets of spam or nonspam. Or, when considering RBLs, having to manually judge whether a given domain is that of a spammer or not.

At the corporate or ISP level, this manual effort is expensive and ongoing. At the individual level, some, like Paul Graham, who suggested using Bayesians, are willing and able to take the time to retrain their Bayesians. But most users are not willing to continuously do this.

One portion of the book is already crucially outdated. In 2004, AOL, and possibly other ISPs, started applying an RBL against domains found inside the bodies of messages, and not just against domains from the headers. As far as I can tell on a reading of this book, whilst it describes various means of finding and applying RBLs, it does not mention the important idea of using them against body domains.