Very true, nice to see someone who knows a bit about Tor. Website fingerprinting attacks analyze patterns in encrypted traffic, looking for preidentified fingerprints associated with certain websites. This sort of attack may be used to identify encrypted Tor traffic with 60% accuracy. This doesn't mean the attacker can tell the difference between your upload of the word dog and the other posters upload of the word cat to some forum, it just means that they can with about 60% accuracy determine that you sent something to that forum. Of course this assumes they are only doing website fingerprinting attack and are monitoring traffic at your entry guard / infrastructure, if they do other things they can of course learn other things. Also I don't think the traffic classifier that CCC used against Tor (getting the 60% accuracy results) used hidden markov modeling (which takes into account not only the fingerprint of a single page, but the multiple possible fingerprints created by browsing through networks of interlinked pages), if it did the accuracy would probably be substantially higher. Using a private (or public via the bridge distribution mechanisms) bridge node that you run yourself is a great way to majorly increase the anonymity offered by the Tor network. For one, you will never be traced by an active attack unless whoever your attacker is takes control of your bridge node somehow (and knows how to target it in the first place). If you control the first hop you can feel a hell of a lot safer about using Tor.