New Browser Fingerprinting Attack on Tor Encrypted Traffic

Why Target Users on the Dark Web?

It might seem odd to target the small pool of users who are on the dark web via the Tor web browser. The special tools and protocols needed to access dark web content make just about any hack more complex.

See, the dark web uses something called 'onion routing' to make privacy and encryption more effective. The user can pick a set of nodes from various countries and territories throughout the world, as advertised by a directory node. Then the Onion routing protocol plays a high-tech game of telephone. Each requested packet is passed from one node to the next, and each message is only decryptable by the next node in the chain. This means that the third hop on the Onion route has absolutely no knowledge of the first gateway, so it can't give a point of origin even if it tries.

So why bother to break the security of a small pool of users with a higher level of paranoia on average?

The simple answer is blackmail. The dark web is a brand new world with a different set of rules. Domains are created in an entirely different way, laws and international policies are often disregarded, and anonymity is held at a premium.

So privacy means everything on the dark web. And the people with the most to hide are generally the most vulnerable if they're discovered. Particularly when the attacker knows that they already have access to a nearly untraceable currency like crypto.

Why Is Tor Even A Thing?

The truth is, there's a need for truly private searching and browsing, and it goes beyond what happens in shady back alleys in cities around the world. Given the record levels of government surveillance around the world, there needs to be a place where those who wish to organize peaceful protests and resistance can meet, discuss their plans, and fund their endeavors. There also needs to be a place where those in 'Big Brother' surveillance states can escape those watchful eyes.

Tor's mission statement on the topic of privacy is simple:

"Privacy is about protecting what makes us humans: our day-to-day behavior, our personality, our fears, our relationships, and our vulnerabilities. Everyone deserves privacy.

The United Nations codified that in the Universal Declaration of Human Rights in 1948. Article 12 states that, “no one shall be subjected to arbitrary interference with their privacy” and that “everyone has the right to the protection of the law against such interference or attacks.” However, governments, corporations, and other powerful entities block us from exercising our right to privacy in many different ways. With commercially available spyware, covert monitoring of our communications, ads that track us around the internet, pseudo-anonymous data sets purchased and sold and used to manipulate sentiment and sow division. And despite all this, we still have to fight to defend and fight to exercise.'

And that's why Tor is here—to help you exercise your human right to privacy even when it's not easy.

Every day, the Tor network helps millions of people connect to the private, uncensored internet. In the face of declining internet freedom, the creep of oppressive and repressive governments, the breakneck advance of surveillance technology, Tor remains a gold standard in privacy and censorship circumvention technology. Our decentralized network, our open-source approach, our community of volunteers, and our commitment to the human right to privacy mean that Tor can offer privacy in a way that few other tools can."

Now that you understand the stakes, let's talk about the first attack of its kind to bring the threat of browser fingerprinting to Onion routing.

About The Fingerprinting Attack

For normal dark web traffic, Tor's Onion routing still does what it always has in the past. It has over six thousand relays to pass your traffic through. There's no direct communication between the entry node and the exit node, so there's no chance that your true route will be discovered.

However, a web browser fingerprinting attack doesn't need to know your route. It can identify you based on a number of parameters asked by the scripts on the website you're trying to visit. Each individual query is innocent by itself: How big is your screen so that we can display things properly? What kind of audio drivers do you have so that we can play the right sounds? What version of the web browser are you using so that we don't try to do anything that isn't supported? Are you on a mobile device that uses things like an accelerometer or gyroscope? What are your keyboard layout and default language?

But when you add up all of those tiny queries, something interesting happens. Your web browser's individual answers can add up to uniqueness. In other words, nobody else in the world has your specific combinations of hardware, software, drivers, and the like.

What are the odds of you having a unique set of web browser parameters? Pretty good, as it turns out. Most people who are taking no precautions will present a unique web browser fingerprint. You can test it out yourself at AmIUnique.

But how would a malicious actor even get access to your Tor traffic? Surely it isn't being snooped on the wire, and even if it was, it's encrypted. It should be quite difficult, to say the least.

Unless of course, the malicious actor is your exit node. By volunteering to be your exit node, a certain amount of information about your browser is available to the exit node 'in the clear'. This compromised exit node will function just like any other, except it will run an analysis on the diversity of traffic generated by the various browsers it encounters. Then, using machine learning, it will try to do an analysis of the traffic.

This is where this new real-world attack differs from the theoretical Tor web browser fingerprinting attacks of the past. This one isn't imposing unrealistic restrictions on users by only letting them visit the 'correct' target sites. Instead, it trains the AI to sift through a huge amount of traffic running through the exit node and identify very specific targets. This is real traffic, not synthetically generated browser sessions run by machine learning. And that makes a lot of difference when it comes to formulating an attack that works in the real world. From the paper:

“Our comparison between static and online models shows that there is an advantage in training on heterogeneous, dynamic traffic when the goal is to fingerprint websites in the wild.”

Because it needs large sample sizes, the exit node will automatically restrict its analysis to the top five percent of sites on the dark web. And ultimately, it will hone in on one to five hugely popular websites that it passes a lot of traffic. According to the paper:

“By studying website fingerprinting (WF) under realistic conditions, we demonstrate that an adversary can achieve a WF classification accuracy of above 95% when monitoring a small set of 5 popular websites.”

The result is an Onion routing exit node, which to the rest of the world looks like it's doing its job, being able to find certain unique web browser fingerprints as users visit the most popular sites on the dark web.

Research is still in the early stages, but given that this was accomplished using real traffic, the threat is certainly viable. These kinds of attacks can only become more accurate as the AI learning model hones in on its best targets. Remember that this is the analysis of a single exit node in a fairly brief time. A larger scope, more computing power, and more time are all that an attack like this takes to reach the next level of significance.

Does This Apply To Normal Web Traffic?

It doesn't need to. Normal web traffic, not using Onion routing at all, is already incredibly susceptible to web browser fingerprinting.

The problem is that many web browsers shut off website functionality to scramble your browser fingerprint. In other words, turning on their fingerprint security features breaks a lot of websites. So to have a 'normal' web browsing experience, their fingerprint mitigation has to be turned down to ineffective levels.

Most users don't have any browser fingerprinting protection turned on unless their browser does so by default. Why? Because most users are simply unaware of the threat that it represents to their privacy. We know this because even in a survey about third-party cookies, a security concern that is going away in 2022 for all major web browsers, people didn't understand the stakes. Less than a third of users adjusted their settings to give themselves more privacy. 9% of users know about and understand anonymous browsing.

So to expect users to investigate web browser fingerprinting specifically, go into their browser settings to maximize their protection, and remain vigilant against future developments is a fairy tale in most cases.

Summing Up

This new web browser fingerprint attack against Tor is certainly disturbing. But it is highly specific, at least at the moment. There's time to make use of privacy tools like Hoody and protect yourself against this and future versions of these attacks that still manage to work over the Onion protocol.