Latest / CyberPodcast / De-anonymizing Anonymous Networks: Attacks & Defenses | Paper Breakdown #10 | CyberPodcast

Transcript

De-anonymizing Anonymous Networks: Attacks & Defenses | Paper Breakdown #10 | CyberPodcast · CyberPodcast

• 0:00
  Welcome to this explainer. I am incredibly excited to guide
• 0:03
  you through this today because we are diving head first into a
• 0:06
  massive invisible cat and mouse game.
• 0:09
  It's happening right now, literally as we speak, all
• 0:12
  across the Internet. It's this high stakes, ongoing
• 0:15
  battle between network defenders, the folks trying to
• 0:18
  protect your privacy, and adversaries who are constantly
• 0:21
  trying to track your data and deanonymize your web traffic.
• 0:25
  Now, if you're wondering why this matters so much, we really
• 0:27
  just have to look at the numbers.
• 0:28
  Recent data shows that encrypted traffic has absolutely exploded.
• 0:32
  It's quadrupled in regions like Europe and Latin America.
• 0:35
  People clearly want privacy, right?
• 0:36
  But here is the chilling reality.
• 0:39
  Simply flipping on encryption. It's no longer enough to hide
• 0:41
  your digital footprint. As we'll see from the research,
• 0:44
  the illusion of total anonymity is kind of fading away, and it's
• 0:46
  setting the stage for this unprecedented privacy arms race.
• 0:50
  O to wrap our heads around this invisible war.
• 0:52
  We'll be covering the set, the battlefield, endtoend attacks,
• 0:56
  single end attacks, the defense, and finally the future.
• 1:00
  Let's get right into it. Section 1 The set OK, so when we
• 1:05
  look at how we currently try to hide our data, we're usually
• 1:07
  talking about flow based anonymous systems.
• 1:09
  Take a single hop system like your standard everyday VPN.
• 1:12
  It basically acts like a direct secure pipe.
• 1:15
  It's super fast, sure, but it has a massive single point of
• 1:18
  failure. If an adversary breaks into that
• 1:20
  one server or that one pipe, your entire data stream is
• 1:23
  exposed. That's exactly why engineers
• 1:25
  built multi hop systems like Tor instead of a single pipe.
• 1:29
  These prioritize high security by sending your data bouncing
• 1:32
  through a complex maze of multiple nodes.
• 1:34
  It cost you some speed, but it buys you a whole lot of
• 1:37
  security. Let's actually breakdown the
• 1:39
  mechanics of how this multi hop architecture works under the
• 1:42
  hood. When you use Tor, your local
• 1:44
  software, which is called the Onion Proxy, packs your data up
• 1:48
  into these uniform little units called cells.
• 1:50
  It then builds a circuit using source routing through three
• 1:53
  distinct stops, an entry node, a middle node, and an exit node.
• 1:57
  The brilliant part of this design, and the reason it's
• 2:00
  leaps and bounds more secure than a standard VPN, is that no
• 2:03
  single router in this entire chain knows both who you are and
• 2:06
  what you're actually doing. Think about it.
• 2:09
  The entry node knows your IP address, but not your final
• 2:11
  destination. And the exit node?
• 2:13
  It knows your destination but has absolutely no idea what your
• 2:16
  IP is. It's a really clever setup.
• 2:19
  Which brings us to Section 3, end to end attacks.
• 2:23
  You see, adversaries aren't just sitting blindly in the dark.
• 2:26
  If they control devices at both the sender and receiver sides,
• 2:29
  they can launch end to end attacks, and these generally
• 2:32
  fall into two buckets. First, attackers can take a
• 2:35
  passive approach, silently monitoring traffic statistics
• 2:39
  like packet counts and timing to correlate patterns between you
• 2:42
  and the server. But because passive monitoring
• 2:45
  is messy and leads to a lot of false positives, they often
• 2:48
  crank things up a notch. They turn to active attacks,
• 2:51
  where they physically inject signals or manipulate traffic to
• 2:54
  confirm a connection with incredibly high accuracy.
• 2:57
  This kind of active manipulation brings us to a truly
• 2:59
  fascinating, if a bit scary, concept, watermarking based
• 3:03
  attacks. The idea here is that an
• 3:05
  adversary embeds a digital watermark.
• 3:07
  You can think of it basically like an invisible ink into the
• 3:10
  network's plumbing right at one end.
• 3:12
  Then they just sit and monitor for that exact same invisible
• 3:15
  ink to emerge at the other end. If they see it pop out, well,
• 3:19
  they have definitively confirmed exactly who is communicating
• 3:22
  with whom. It's a highly targeted, highly
• 3:24
  effective technique. So you might be wondering, how
• 3:28
  exactly do you put invisible ink into network traffic?
• 3:31
  Well, at the network layer adversaries achieve this through
• 3:33
  some seriously technical manipulation.
• 3:35
  They might alter the traffic rate to embed spread spectrum
• 3:38
  signals, or subtly manipulate packet delay intervals.
• 3:41
  Imagine a train schedule. If an attacker subtly delays
• 3:44
  every 3rd train, they create a totally unique rhythm, right?
• 3:47
  By shifting these arrival times, they're literally encoding
• 3:50
  invisible binary ones and zeros into the flow of your data.
• 3:54
  They can also do this by carefully altering specific
• 3:56
  packet sizes, mapping packet links to hex bits, all while
• 3:59
  maintaining the overall traffic distribution so the interference
• 4:02
  remains completely 100% invisible to you.
• 4:06
  And honestly, it goes even deeper than that, right down to
• 4:08
  targeting the protocol layer of Tor itself.
• 4:11
  See, Tor uses an encryption standard called AESCTR.
• 4:15
  This relies on a synchronized encryption counter.
• 4:18
  It's kind of like 2 spies reading from the exact same page
• 4:21
  of a secret code book. If an attacker sitting at the
• 4:23
  entry node silently deletes, inserts, or replays a single
• 4:27
  data cell, the encryption counter desynchronizes.
• 4:30
  Suddenly the rest of the nodes are reading the wrong page of
• 4:32
  the code book. When your traffic finally hits
• 4:34
  the exit node, the decryption intentionally fails.
• 4:37
  And that failure, it serves as a perfect silent signal back to
• 4:40
  the attacker, confirming your identity.
• 4:42
  It's just wild how clever that is.
• 4:45
  All right, moving on to Section 4.
• 4:47
  Single and fingerprinting attacks.
• 4:50
  Now what if an attacker can't control both ends of the
• 4:52
  network? Well, they can still launch
• 4:53
  single end attacks, and the most prominent of these is called
• 4:56
  website fingerprinting or WF. This is a passive single end
• 5:01
  attack. Remember, even though your
• 5:02
  traffic is perfectly encrypted, the unique patterns of that
• 5:05
  traffic, the exact sizes of the packets, the timing of the data
• 5:08
  bursts, they actually act just like a human fingerprint.
• 5:11
  The adversary simply matches the unique traffic patterns of your
• 5:14
  encrypted connection against a massive pre collected database
• 5:18
  of known website traffic characteristics.
• 5:20
  Boom, they know exactly what you're looking at.
• 5:23
  To actually build this out in the real world, the adversary
• 5:25
  follows a few steps. First, they do offline training.
• 5:29
  They literally emulate normal browsing on target websites just
• 5:32
  to build up that fingerprint database.
• 5:34
  Second, they preprocess this traffic to filter out the noise.
• 5:38
  You know things like dynamic changing ads that might mess up
• 5:40
  the data. Third, they silently deploy a
• 5:43
  monitor on the victims local proxy link.
• 5:46
  And finally, step four, they perform online classification.
• 5:49
  They use machine learning algorithms to match the live
• 5:52
  traffic features they are seeing to the fingerprints in their
• 5:54
  database. It is entirely passive, yet it
• 5:56
  completely exposes your browsing habits.
• 5:59
  But attackers can also go on the offensive on a single end.
• 6:02
  In single end active attacks, if an adversary, say controls the
• 6:06
  Tor exit node, they are sitting right on that unencrypted link
• 6:09
  between the proxy and the destination server.
• 6:11
  From that vantage point they can actively inject malicious code,
• 6:15
  things like JavaScript, Flash, or Java right into the non
• 6:18
  encrypted traffic on the server side.
• 6:20
  When your browser blindly executes that code, it forces a
• 6:23
  direct connection, completely bypassing your proxy settings
• 6:26
  and instantly revealing your real IP address back to the It's
• 6:29
  a total end run around your security Section 5
• 6:33
  countermeasures and defense strategies.
• 6:36
  OK, take a breath, because it's not all doom and gloom.
• 6:39
  As terrifying as these attacks sound, network defenders are
• 6:43
  constantly engineering brilliant countermeasures to protect your
• 6:46
  data across multiple layers. The goal?
• 6:49
  Destroy the fingerprints. At the network layer, they use
• 6:52
  packet padding. They add dummy data so every
• 6:55
  single packet is puffed up to the maximum transmission unit,
• 6:58
  or MTU. It's basically like making every
• 7:00
  single package you ever mail look identical, whether it holds
• 7:03
  a bowling ball or a single feather.
• 7:05
  They also inject dummy traffic. At the protocol layer we see
• 7:09
  cell padding, which is heavily used in systems like SSH and
• 7:12
  TLS, and up at the application layer they use tricks like
• 7:15
  Http://pipelining, or even intentionally load invisible
• 7:18
  background decoy web pages while you browse.
• 7:21
  This completely scrambles the traffic bursts and totally
• 7:24
  breaks the attacker's machine learning classifiers.
• 7:27
  Honestly, one of the most fascinating defenses detailed in
• 7:30
  this explainer is a network layer technique called traffic
• 7:32
  morphing. Instead of just randomly
• 7:34
  scrambling data, this defense intentionally varies your
• 7:37
  current traffic patterns to mimic the exact packet size
• 7:40
  distribution of a totally different web page.
• 7:43
  So you could be reading a highly sensitive, confidential news
• 7:46
  article, But the morphing algorithm meticulously shapes
• 7:49
  your encrypted packets so that to any attacker monitoring the
• 7:51
  line, it mathematically looks exactly like you're browsing a
• 7:54
  generic search engine. Or, you know, scrolling through
• 7:57
  pictures of cats. It's absolute genius.
• 8:00
  And finally, Section 6, the ongoing privacy arms race.
• 8:05
  Ultimately, no single defense is perfect.
• 8:07
  It's a perpetual cycle. To survive this ongoing arms
• 8:10
  race, the source material emphasizes a real need for
• 8:12
  theoretical channel modelling. What does that mean in plain
• 8:15
  English? It means we have to understand
• 8:17
  the interactions between these attacks and countermeasures by
• 8:19
  treating them as binary messages sent through a noisy
• 8:21
  communication channel. Basically, treating it like
• 8:24
  static on a radio. By applying hardcore
• 8:26
  communication theory, computer scientists can calculate the
• 8:29
  exact mathematical limits of our privacy and design systems that
• 8:32
  are structurally sound against future unknown attacks.
• 8:35
  So, having explored the incredible technical depths of
• 8:38
  this invisible war raging right beneath our web browsers every
• 8:41
  single day, we are left with a truly vital question to ponder
• 8:45
  as attackers develop even more active, sophisticated ways to
• 8:48
  manipulate and fingerprint our traffic.
• 8:50
  Is true online anonymity just a solvable math problem waiting
• 8:54
  for the right defence? Or is it merely a temporary
• 8:56
  illusion waiting to be broken? Thank you so much for joining me
• 8:59
  for this explainer. I really hope this gave you a
• 9:01
  powerful new perspective on exactly how your data moves
• 9:04
  across the web.