We present a Federated Learning (FL) based solution for building a distributed classifier capable of detecting URLs containing sensitive content, i.e., content related to categories such as health, political beliefs, sexual orientation, etc. Although such a classifier addresses the limitations of previous offline/centralised classifiers, it is still vulnerable to poisoning attacks from malicious users that may attempt to reduce the accuracy for benign users by disseminating faulty model updates. To guard against this, we develop a robust aggregation scheme based on subjective logic and residual-based attack detection. Employing a combination of theoretical analysis, trace-driven simulation, as well as experimental validation with a prototype and real users, we show that our classifier can detect sensitive content with high accuracy, learn new labels fast, and remain robust in view of poisoning attacks from malicious users, as well as imperfect input from non-malicious ones.@en

In recent years, governments worldwide have moved their services online to better serve their citizens. Benefits aside, this choice increases the danger of tracking via such sites. This is of great concern as governmental websites increasingly become the only interaction point with the government. In this paper, we investigate popular governmental websites across different countries and assess to what extent the visits to these sites are tracked by third-parties. Our results show that, unfortunately, tracking is a serious concern, as in some countries up to 90% of these websites create cookies of third-party trackers without any consent from users. Non-session cookies, that are created by trackers and can last for days or months, are widely present even in countries with strict user privacy laws. We also show that the above is a problem for official websites of international organizations and popular websites that inform the public about the COVID-19 pandemic.

@en