Is Bayesian Censorware possible?

There has been some talk on this forum and others about Bayesian spam filtering. People seem to be saying positive things about it.

That begs the question as to how useful Bayesian analysis might be in blocking "inapropriate content" or identifing possible terrorist communications, etc.

Looking at how well it filters spam I suspect it would be very effective at censorware, as it is essentially the same task.

It would be hopeless for spotting terrorist comms, at least if I was one of the terrorists, because I would have encrypted the message.