Latest / CyberPodcast / How Tor Guard Selection Can Be Exploited | Paper Breakdown #3 | CyberPodcast

Transcript

How Tor Guard Selection Can Be Exploited | Paper Breakdown #3 | CyberPodcast · CyberPodcast

• 0:00
  All right, let's jump right into this explainer.
• 0:02
  Today we're looking at a completely fascinating academic
• 0:05
  paper that outlines a brilliant and honestly pretty terrifying
• 0:09
  stealth attack on one of the world's most popular anonymity
• 0:12
  networks. We're pulling from research
• 0:14
  published in the International Journal of Security and its
• 0:16
  Applications. And I got to say, it reads way
• 0:19
  less like a dry academic paper and a lot more like a blueprint
• 0:23
  for a high tech digital heist. So right now there are nearly
• 0:27
  2.5 million users relying on the Tor network.
• 0:30
  They're using it because they believe it offers this
• 0:33
  unbreakable shield against surveillance, censorship,
• 0:36
  tracking, the whole shebang. It's basically the gold standard
• 0:39
  for online privacy. But when we actually look at the
• 0:42
  data presented in this research, we have to ask a really tough
• 0:46
  question. Is that guarantee of absolute
• 0:49
  anonymity actually a myth? 30?
• 0:52
  Yep, just 30 minutes. That is the shockingly brief
• 0:55
  amount of time it takes for a coordinated attacker to
• 0:58
  completely compromise a specific Tor users entry point.
• 1:01
  Literally just half an hour to pick the digital lock and
• 1:04
  severely degrade a target's privacy.
• 1:07
  Here is our road map for this explainer.
• 1:09
  First, the Tor anonymity shield. Second, the guard node
• 1:13
  bottleneck. Third, the stealthy attack
• 1:15
  strategy. 4th forcing a new selection, 5th the 30 minute
• 1:20
  compromise and 6th the real world fallout.
• 1:23
  OK, Section 1, the Tor anonymity shield and how this digital
• 1:27
  disguise actually works. When you're using Tor, your data
• 1:30
  doesn't just travel in a straight line from your computer
• 1:33
  to a website. It bounces.
• 1:35
  It goes from the client to an entry node, then a middle node,
• 1:38
  then an exit node, and finally to the server.
• 1:41
  The researchers make a great point here.
• 1:43
  To completely de anonymize a user, an attacker generally
• 1:46
  needs the incredibly rare luck of monitoring both that entering
• 1:49
  segment and the exiting segment at the exact same time.
• 1:52
  It's like trying to have eyes on both the entrance and the exit
• 1:55
  of a massive maze simultaneously.
• 1:58
  It's designed to be practically possible.
• 2:00
  Now, an attacker generally has two ways to try and break this.
• 2:04
  Over on the left, you've got application layer attacks.
• 2:07
  This is where they try to hack the specific browser software,
• 2:10
  maybe using JavaScript vulnerabilities or something
• 2:12
  similar. It's highly accurate, sure, but
• 2:15
  it's very specific. But look at the right side.
• 2:17
  This is what's called a traffic confirmation attack, and it is
• 2:20
  way more reliable. The attacker doesn't even need
• 2:22
  to hack your browser. They literally just watch the
• 2:24
  network frequency and time patterns at both the entry and
• 2:27
  exit points. If the signal going in matches
• 2:30
  the rhythm of the signal coming out, boom, they've got you
• 2:33
  moving on to Section 2. The guard node bottleneck
• 2:37
  basically the vault door to your data.
• 2:40
  So that fault door is called the entry guard node.
• 2:43
  You can really think of the entry guard as a highly loyal
• 2:46
  digital bouncer. Now, back in the early days of
• 2:48
  Tor, your entry point changed constantly, but that meant
• 2:51
  eventually, just by pure mathematical probability, you
• 2:54
  would randomly pick an attacker's node.
• 2:57
  So Tor changed the rules. Now your system picks a guard
• 2:59
  node and sticks with it for 30 to 60 days because you keep the
• 3:03
  exact same bouncer for up to two months and attacker usually has
• 3:06
  to wait a very very long time just hoping for a chance to
• 3:08
  compromise your connection. But here is the double edged
• 3:12
  sword of that exact system. Yes, sticking with one guard
• 3:15
  reduces the risk of randomly hitting a bad one overtime.
• 3:18
  However, if an attacker can somehow force you to pick a
• 3:21
  compromised bouncer, well you are locked into using their
• 3:24
  corrupted entry point for the next month or two.
• 3:27
  Literally every single circuit you build will start with them.
• 3:30
  Which brings us to Section 3, the stealthy attack strategy, or
• 3:35
  how they plan the heist. The researchers lay out this
• 3:38
  beautifully elegant 4 step lock picking method.
• 3:41
  Step 1. The attacker passively monitors
• 3:44
  the user's traffic just to identify their current guards.
• 3:46
  Step 2. They actively block connections
• 3:48
  to those specific normal guards. Step 3.
• 3:51
  This disruption forces the Tor software to automatically seek
• 3:54
  out a new guard. And finally, step four, they
• 3:56
  just repeat this process until the user blindly walks right
• 3:59
  into a malicious guard deployed by the attacker.
• 4:02
  Now you're probably thinking, OK, but who can actually pull
• 4:05
  this off? The paper outlines some very
• 4:07
  specific attacker requirements. First, the attacker needs to
• 4:10
  deploy several high bandwidth guard nodes into the Tor
• 4:13
  network, which by the way, is actually pretty cheap and easy
• 4:16
  to do. But they also need the power to
• 4:18
  passively monitor and actively drop specific local connections
• 4:22
  at the user's end. So this means the adversary has
• 4:25
  to be someone like a local network administrator, maybe a
• 4:27
  corrupt Internet service provider, or even a state level
• 4:30
  actor. Basically, they have to
• 4:32
  completely control the local Rd. leaving your house Section 4,
• 4:36
  forcing a new selection. This is how they actually pick
• 4:40
  the lock. It all really comes down to
• 4:42
  Tor's own user experience design.
• 4:44
  You see, Tor is built for seamless usability.
• 4:47
  If your Internet connection to a guard node suddenly drops or
• 4:49
  fails, your Tor client doesn't just give up and crash.
• 4:52
  No, it automatically and silently seeks out a new entry
• 4:55
  guard from the directory to maintain a fast, reliable
• 4:58
  connection for you. It's just trying to be helpful.
• 5:00
  But by repeatedly blocking the user's legitimate entry nodes,
• 5:03
  the attacker is literally weaponizing Tor's own auto
• 5:06
  reconnect feature. The software blindly cycles
• 5:08
  through options over and over until it lands squarely on the
• 5:11
  attacker's malicious node. And the math behind this attack
• 5:15
  reveals a pretty frightening reality.
• 5:17
  Looking at the relationship between the number of malicious
• 5:20
  guards the attacker deploys into the network on the bottom axis
• 5:23
  and the number of tries or connection blockings needed to
• 5:26
  trap the user on the vertical axis.
• 5:29
  Well, it shows that an attacker doesn't need to control half the
• 5:32
  network. Not even close.
• 5:34
  Even with just a few dozen malicious guards deployed in the
• 5:37
  massive Tor network, it only takes a handful of forced
• 5:40
  connection blockings to spring the trap.
• 5:42
  Once you hit about 30 deployed guards, the number of connection
• 5:45
  drops needed just plummets. All right, Section 5, The 30
• 5:49
  minute compromise. The getaway.
• 5:51
  The researchers didn't just theorize about this, by the way.
• 5:54
  They ran actual simulations using tours path selection
• 5:58
  algorithms. The experimental results show
• 6:00
  what happens when the attacker deploys just 10 high bandwidth
• 6:03
  malicious guard notes. Even with such an incredibly
• 6:06
  small footprint, they can force the user right into their trap
• 6:09
  after just a few dozen interruptions.
• 6:12
  But if the attacker scales up even slightly, the results are
• 6:15
  just staggering. If they deploy 50 high bandwidth
• 6:18
  guards, which is incredibly feasible for a determined
• 6:21
  adversary, they can compromise more than 80% of targeted users
• 6:25
  in less than 20 connection blocks.
• 6:27
  Just think about that. That's an 80% success rate
• 6:30
  achieved by dropping the connection fewer than 20 times.
• 6:33
  Now remember that Tor builds new circuits quickly to maintain
• 6:36
  your Internet speed. In their tests, the researchers
• 6:39
  found the mean interval between guard selection rounds was
• 6:42
  extremely fast when traffic is flowing.
• 6:44
  So those 20 blocked connections? They translate to just 30
• 6:47
  minutes of real world time. As the researchers state
• 6:50
  plainly, an attacker can successfully hijack the user's
• 6:52
  entry guard in about half an hour.
• 6:54
  Just 30 minutes. Finally, Section 6, the real
• 6:58
  world fallout. What this actually means for our
• 7:01
  privacy. For me, the truly frightening
• 7:04
  part of this exploit is it's absolute stealth.
• 7:06
  It operates completely under the radar.
• 7:09
  The network failure looks entirely natural to the Tor
• 7:12
  client. To you it just looks like normal
• 7:14
  Internet churn, or maybe a slightly spotty Wi-Fi
• 7:17
  connection. There is no malware injected, no
• 7:20
  complex encryption broken. It's just clever traffic
• 7:22
  manipulation, and it's a really dark piece of irony that Tor's
• 7:26
  obsessive desire to give you a smooth, uninterrupted browsing
• 7:29
  experience is exactly what dooms you hear.
• 7:32
  Because of Tor's own safety rules, once the victim lands on
• 7:35
  that bad node, they are permanently stuck with it for up
• 7:38
  to two months without ever knowing.
• 7:40
  This gives the attacker a massive, prolonged window to sit
• 7:43
  back, watch the traffic, compromise an exit node, and
• 7:46
  execute a full deanonymization attack.
• 7:49
  This research is honestly a master class in turning a
• 7:52
  system's strengths against itself, and it leaves us with a
• 7:54
  massive critical question for the future of online privacy.
• 7:58
  If the very mechanism is designed to protect our
• 8:00
  anonymity, like the guard node system and auto reconnect
• 8:03
  features can be weaponized against us in just 30 minutes,
• 8:06
  how do we build a truly untraceable network?
• 8:09
  It kind of proves that as long as the digital roads and
• 8:11
  bouncers we rely on can be manipulated by those who control
• 8:14
  the infrastructure, well, our data is never, ever entirely out
• 8:18
  of reach.