The most promising technical attack on spam…

… still seems to be SpamBayes which implements and develops Paul Graham’s ideas for using Bayesian inference to filter messages. More info here.