Spam filtering has been in Thunderbird for a decade now (based on a Bayesian classifier, although it's not quite naïve Bayes). There is a capability to extend the classifier to problems beyond spam/ham classification for extensions.
Interestingly, scam detection (rather than spam filtering) is a simple hardcoded rule, which looks at whether the send headers match up. I've received more than a few genuine emails marked as potential scams because they were sent from a rebranded mail provider.
