Website fingerprinting attacks let attackers determine which websites a user visits, posing a significant risk to online privacy. Website fingerprinting attacks have been demonstrated both in the network-based setting, where the adversary is able to observe the network traffic between the user and the secure network, and in the cache-based setting, where the adversary injects malicious JavaScript code into the user’s browser and observes memory activity. In both of these settings, previous research has demonstrated that state-of-the-art website fingerprinting attacks can succeed even in highly-constrained environments, for example when attacking the privacy-enhanced Tor browser. One known limitation of website fingerprinting attacks, however, is their sensitivity to concept drift — as the time difference between the training period and the actual attack grows, the accuracy of the model used in the fingerprinting attack degrades due to changes in webpage content, network conditions, and the software implementation of the browser. This phenomenon is a known challenge in website fingerprinting. This study provides a quantitative analysis of the effect on website fingerprinting attacks of concept drift in both the network-based and cache-based settings, based on a website fingerprinting trace dataset collected over a period of several months. It examines the effect on accuracy of changes both to the browser version and to the website content. It then investigates multiple approaches for addressing concept drift, assuming the attacker can add a limited amount of fresh data to his training dataset. Our evaluation shows that using advanced machine learning techniques can significantly reduce the effect of concept drift on website fingerprinting attacks, and that, specifically, an incremental learning approach nearly maintains the model’s accuracy over time, while requiring only 2 % of the data needed for full retraining. This finding demonstrates the practicality of long-term website fingerprinting attacks in the real world.

Website fingerprinting attacks use statistical analysis on network traffic to compromise user privacy. The classical attack model used to evaluate website fingerprinting attacks assumes an on-path adversary, who observes traffic traveling between the user's computer and the network. In this article we investigate a different attack model, in which the adversary sends JavaScript code to the target user's computer. This code mounts a cache side-channel attack to identify other websites being browsed. Using machine learning techniques to classify traces of cache activity, we achieve high classification accuracy in both the open-world and the closed-world models. Our attack is more resistant than network-based fingerprinting to the effects of response caching, and resilient both to network-based defenses and to side-channel countermeasures. We carry out a real-world evaluation of several aspects of our attack, exploring the impact of the changes in websites and browsers over time, as well as of the attacker's ability to guess the software and hardware configuration of the target user's computer. To protect against cache-based website fingerprinting, new defense mechanisms must be introduced to privacy-sensitive browsers and websites. We investigate one such mechanism, and show that it reduces the effectiveness of the attack and completely eliminates it when used in the Tor Browser.