3

During a panel at the 2014 SXSW conference, Julian Assange (founder of Wikileaks) stated "The ability to surveil everyone on the planet is almost there and, arguably, will be there in the next couple of years".

If we get to the point where all network traffic was recorded at every ISP, will Tor still be effective at anonymization?

6

The short answer is no. Basically, Tor is susceptible to timing correlation attacks, where someone who is observing the connection going from your client (OP) to the Tor network and also the connection from the Tor network to your destination can tell with fairly high certainty that they are the same connection.

Note that you would need to correlate data from multiple ISPs to do this. The traffic confirmation scenario, where I suspect that user X at ISP A is accessing site Y at ISP B, is fairly easy. The full deanonymization scenario would require me to analyze traffic from a large number of ISPs to look for correlations. As far as I know, there hasn't been an analysis of how easy this is to do at Internet scale, but there's no good reason to believe that it couldn't be done.

Note: For a timing attack to work they need to be able to single out just your traffic on Tor. If you run a full node on your computer and also use Tor to browse the Internet they cannot distinguish the traffic coming from you and for all the other routing taking place. This makes it impossible for them to do this with any reliability.

  • Would it help substantially to work through a hosted workstation? The user would connect to the workstation (hidden service) via RDP/TLS, and the workstation would access the Internet through a separate Tor instance. – mirimir Mar 26 '14 at 4:47
  • 1
    Not substantially. You could use timing correlation to link your activity to the workstation, and then again to link the workstation to your computer. There may even be enough timing correlation to link your computer to your activities directly. – Nikita Borisov Mar 26 '14 at 11:32
  • 1
    I wonder how much latency and jitter would be required to counter timing correlation attacks. It would probably require too much for a pleasant user experience, right? – mirimir Mar 27 '14 at 3:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.