High-Fidelity Provenance: Exploring the Intersection of Provenance and Security

In the past 25 years, the World Wide Web has disrupted the way news are disseminated and consumed. However, the euphoria for the democratization of news publishing was soon followed by scepticism, as a new phenomenon emerged: fake news. With no gatekeepers to vouch for it, the veracity of the information served over the World Wide Web became a major public concern. The Reuters Digital News Report 2020 cites that in at least half of the EU member countries, 50% or more of the population is concerned about online fake news. To help address the problem of trust on information communi- cated over the World Wide Web, it has been proposed to also make available the provenance metadata of the information. Similar to artwork provenance, this would include a detailed track of how the information was created, updated and propagated to produce the result we read, as well as what agents—human or software—were involved in the process. However, keeping track of provenance information is a non-trivial task. Current approaches, are often of limited scope and may require modifying existing applications to also generate provenance information along with thei regular output. This thesis explores how provenance can be automatically tracked in an application-agnostic manner, without having to modify the individual applications. We frame provenance capture as a data flow analysis problem and explore the use of dynamic taint analysis in this context. Our work shows that this appoach improves on the quality of provenance captured compared to traditonal approaches, yielding what we term as high-fidelity provenance. We explore the performance cost of this approach and use deterministic record and replay to bring it down to a more practical level. Furthermore, we create and present the tooling necessary for the expanding the use of using deterministic record and replay for provenance analysis. The thesis concludes with an application of high-fidelity provenance as a tool for state-of-the art offensive security analysis, based on the intuition that software too can be misguided by "fake news". This demonstrates that the potential uses of high-fidelity provenance for security extend beyond traditional forensics analysis