Privacy Uitdagingen in het Tijdperk van Grootschalige Dataverzameling

As daily life becomes increasingly digitized, service providers and data brokers collect, store, and process massive amounts of personal data. People share large portions of their activities and habits online, releasing extensive trails of sensitive personal information. Simultaneously, mobile applications and online services further contribute to the large-scale collection, storage, and processing of personal data. Given its high value for intelligence and commercial purposes, this data is increasingly traded between parties to enable the creation and enhance the accuracy of machine learning models. This raises significant privacy concerns, as individuals often have little to no control over how their personal information is used after collection. Data brokers aggregate vast amounts of data and sell it to third parties, often without caring about the legitimacy of the intended use. Data de-identification offers a potential solution to enhance user privacy. However, past studies have shown that careless anonymization is largely ineffective. A major reason is that the privacy level of de-identified datasets is often measured in isolation. However, in real-world settings, the link between de-identified data and individuals can often be reconstructed by enriching it with publicly available information. Note further that data breaches are becoming increasingly frequent. A notable example is the Gravy Analytics data breach of January 2025, which exposed precise location traces of millions of individuals. This is just one concrete case in which large volumes of personal information have been leaked online, severely compromising the privacy of individuals. However, many attack vectors exist. Malicious actors can crawl public data sources, exploit leaks, or purchase large-scale datasets containing detailed personal data. Using this information, they may inflict direct harm through blackmail, extortion, coercion, or hate crimes, or employ the acquired data to re-identify individuals. This PhD dissertation investigates the risks and impact associated with the careless collection, storage, and release of sensitive personal data. Through a series of large-scale experiments on various (semi-)public datasets, the research identifies and assesses potential threats to the privacy of individuals represented in such data. Amongst others, this thesis conducts an in-depth analysis of a large collection of user reviews posted on Google Maps, as well as a study of a large-scale dataset containing location traces of millions of individuals. The findings aim to raise awareness towards both individuals and policymakers about the real-world implications of data exposure. Furthermore, this dissertation investigates the strengths and limitations of data de-identification techniques as a potential technical approach to the risks associated with managing personal data. It demonstrates the viability of data anonymization techniques in enhancing privacy when releasing data, without significantly compromising its utility. However, this work also demonstrates that residual privacy risks persist even after thorough anonymization, particularly when multiple versions of the same dataset are released. Lastly, based on the findings, this dissertation proposes guidelines for various stakeholders to support privacy-friendly data management practices.