By default, Tor doesn’t protect you from nation states. It’s a start, but you have to be an intelligent user who understands statistics to have some protection from nation states.

Let’s assume there’s two teams, because in geopolitics, it seems like we divide into “west” and “east.” Let’s assume team 1 controls 10% [1] of the relays, they have more than enough budget to pay for the entire network 100x over. That means, on entry, there’s a 10% probability that you will land on their entry node.

Now, to do traffic analysis, they need you to also land on their exit. The probability of that is also 10% in the example. In other words, 10% of the time that you have their entry, you will also have their exit. (or, for 1 in every 100 circuits, you will have a compromised circuit) If you use Tor everyday for a year, you’ll likely have a fucked circuit at least once. If you use something like Whonix that spawns like 10-20 circuits at start, you’ll have a compromised circuit weekly.

A compromised circuit isn’t the end of the world. Most internet traffic today uses end to end encryption, [2] so as long as the service is outside of team 1’s jurisdiction, your communications are safe… but team 1 knows who you are, and that you are talking to someone they don’t trust. If it’s in their jurisdiction, they can get a warrant, and they can fully de-anonymize the traffic between the service that you were using.

All of this is to say, it’s hard to stay in the dark if your adversary is information intelligence. The best way to stay invisible is to use the network as infrequently as possible, and to make the time correlation very far off. (Use custom relays that delay when the traffic travels so that traffic analysis like this example is not possible)

By the way, in the US, the NSA has multiple sites where they copy the traffic on the backbone for analysis. They’re performing some deep packet analysis. These systems are going to improve in the future with machine learning. As an example, in China, it’s not exactly simple to connect to Tor as some methods of concealing Tor traffic result in detection from machine learning that they’re performing on all traffic.

[1] This is a hypothetical. They could control 0%, 5%, 25%, etc. It’s publicly unknown how much they control or if they try to control the network at all.

[2] Be careful with your assumptions about https. Where are the root authorities? Why should we trust them? It’s better security to never trust them.