The modern digital economy relies on continuous data exchange, but existing foundational mechanisms for data protection primarily govern data at rest. Once an authorized agent retrieves data, the semantic context of its intended purpose is lost, making this data-centric paradigm insufficient for articulating and addressing today's complex data problems. This dissertation argues for a shift toward governing data in motion through a dataflow-centric paradigm. We introduce the dataflow preferences framework that enables agents to articulate enforceable preferences over dataflows, which are abstractions over the movement of data among agents. To enforce these preferences within the application ecosystem, we propose a bolt-on data escrow architecture. Rather than pulling raw data to centralized platforms, this model leverages delegated computation, requiring applications to send their computational tasks to a trustworthy intermediary operating within the individual's trust zone. For statistical computations, Differential Privacy (DP) serves as a rigorous algorithmic intervention to enforce preferences on the computation function of a dataflow. To address the practical difficulty data controllers face when choosing the privacy parameter $\epsilon$ in DP, we propose solutions to help controllers make more informed choices of $\epsilon$, thus enabling practical deployment of DP intervention. Together, the dataflow preferences framework and the proposed technical interventions provide the tool to articulate data problems and concrete mechanisms to incentivize, block, or modify dataflows, ensuring that agents' dataflow preferences are effectively enforced.
