I believe they're referring to the traffic analysis style attacks. You don't have to know the content of the traffic if you control the entry node and exit node

I believe they're referring to the traffic analysis style attacks. You don't have to know the content of the traffic if you control the entry node and exit node, either the guard relay, or the hidden service by monitoring the size and timing of individual packets. This doesn't deanonymize on its own unless you control the hidden service, but can be used with other data to link anonymous identities to the IP connecting to the entry node. That being said, what they're referring to is pretty old research and I haven't heard about any evidence proving any agency controls a large amount of the network. For a more current account of modern attack methods against Tor this paper[1] is a pretty recent compilation of researched attacks. The Tor developers actively try to defend against these but you should still be careful, it isn't researched enough yet (read: Tor isn't that old) to defend against resource rich attackers. [1]: https://arxiv.org/pdf/1803.02816.pdf https://arxiv.org/pdf/1803.02816.pdf