[iamnothere]

It’s extremely unlikely that they would be able to find an end user (not an onion site operator, a user) with good opsec who connects occasionally, such as a journalist uploading a few documents to a secure onion drop. All existing known attacks were against onion site operators running for long periods from a static location (still took a lot of resources and time to track them down) or end users with poor opsec/infosec.

The whole thing reads as scaremongering FUD to prevent people from using Tor, with further FUD tacked on to make people think that using it might be illegal somehow. Tor is actually great for personal infrastructure (no need for domain names or a static IP), limited anonymity, and censorship resistance.

[neutered_knot]

The site linked takes a shot at enumerating how unlikely it is. Do you claim it is wrong? If so, what is your calculated chance?

To me, TOR is not adequate to protect users targeted by a nation state who are the ones that TOR claims to be created for.

[iamnothere]

Given that onion sites require six hops and that the Tor team keeps watch for suspicious node behavior, and that there is no “exit node” where you can more closely observe outgoing traffic, onion connections are actually very tough to correlate. It requires a large number of compromised nodes plus cooperation with ISPs and backbone providers, as seen in the arrest of the onion site operator a few years ago. There were some good writeups at the time. You basically need to use DDoS techniques combined with targeted disconnections to narrow down the list of potential targets, even while owning many nodes. And onions have seen some DDoS hardening since this time.

Clearnet traffic via exit node is a bit different. With only three hops it might be possible to correlate targeted traffic by owning a huge number of nodes, but even then, unless you also control the server being connected (or it barely receives any traffic) then it may not give you anything actionable. (Using Tor is not a crime.) Unless you can see what is being done on the server by the unmasked user, or you can establish a pattern of behavior, or you see something like a large data transfer whose size matches a known event of interest, then all you know is someone accessed the server over Tor. And even then, owning both the entry and exit isn’t sufficient if the user is masking their traffic with decoy and/or relay traffic.

[neutered_knot]

But you don't need all the hops. You can run a correlation attack (which has been long known):

https://github.com/Attacks-on-Tor/Attacks-on-Tor

and if you can get the guard and exit node for a clearnet connection and the guard, rendezvous point and exit for the onion service that can be enough.

[iamnothere]

You ignored a substantial portion of the reply. “That can be enough”... yet it has not been. It’s actually very difficult to perform correlation attacks in a complex network, especially if the user is generating decoy traffic or passing along relay traffic, and even moreso if the end server is highly active as well. It takes an enormous amount of resources to even determine that someone may be connecting to a specific server, and again just connecting to a specific server often tells you nothing about what the user was specifically doing. So unless you’re hunting down someone selling enriched uranium, major abuse content producers/hosters, or something of that scale, putting in all that investment to gain a fuzzy data point that likely isn’t even useful in an enforcement context is just not worth the tradeoff.

Come back when you have evidence of real-world attacks and not just FUD against the best current network for anonymity.

[neutered_knot]

> So unless you’re hunting down someone selling enriched uranium, major abuse content producers/hosters, or something of that scale, putting in all that investment to gain a fuzzy data point that likely isn’t even useful in an enforcement context is just not worth the tradeoff.

But I don’t think we disagree. My view is that TOR is inadequate against a nation state attack because for some of these attacks it is easier to do mass de-anonymization and hope you get some particular user or set of users you are interested in. The resources to do this are small for something the scale of an intelligence agency, but excessively large for some local police department.

I’m not sure why you appear so hostile to citing attacks that are well-known and already part of the public threat model.

[iamnothere]

Even mass deanonymization—which is unlikely to be sustainable long-term in such an adversarial ecosystem—does nothing for you unless you (a) know specifically what you are looking for and (b) are able to either compromise the end server or identify frequent or very large transfers that correspond exactly with known events or data of interest (like specific uploads). There are some countries where the authorities might be allowed to round up anyone who has connected to a server without further information about what they were doing, but despite our declining civil liberties situation, these countries aren’t currently in the West.

There just aren’t that many people who are both legitimate and likely targets of such an attack. And since the most likely actor to be able to afford such an attack (USG) also has practical uses for Tor, IMHO it would be unlikely to do anything that actually threatens the network. I could be misremembering, but I believe the one big successful deanonymization attack was in Europe, not the US, and the approach used there would not have worked to locate an occasional end user of a busy server.

I am not really interested in debating this further. Feel free to respond of course, but it’s obvious to me (and hopefully everyone else) that you have an axe to grind against Tor.