Intellectual Property Protection in Open Data Sharing

The dissertation addresses the challenge of protecting data ownership and intellectual property (IP) in shared digital environments through technical safeguards. Its goal is to ensure that data owners retain control even after sharing by embedding verifiable, recipient-specific marks into datasets. This enables traceability and ownership verification in cases of unauthorised redistribution. At the core of this work is the development of an open-source framework that includes implementations of well-known techniques for protecting data ownership and can be integrated into existing data-driven platforms and extended with novel methods, attacks and measures.

By enabling enforceable digital rights, this work empowers data contributors, lowers the risk of misuse, and encourages open and accountable data sharing. This supports fair, transparent, and trustworthy internet and contributes to broader societal goals like digital sovereignty and ethical data governance.

Data ownership protection techniques, such as watermarking and fingerprinting, have been widely studied in the context of digital rights management and applied across various forms of digital content, such as multimedia and ML models (survey by Barni et al.). However, less attention has been given to structured datasets due to limited embedding surfaces and various technical challenges. The most recent approaches build on the early techniques that use pseudo-random embedding into the leastsignificant bits of data, with stronger fidelity measures, such as correlation preservation and downstream utility. We first addressed this in 2020 with our correlation-preserving fingerprinting method. In 2023, Ji et al. formalised correlation attacks. Zhang et al. proposed a method that, besides ownership verification, also enables privacy preservation. Despite these advances, many existing solutions lack robustness, generalisability, or blind detection capabilities.