Sharing data to enable artificial intelligence (AI) development creates unique tensions in contemporary data governance: balancing model robustness and bias reduction with strict requirements to protect privacy and personal rights, defining conditions for responsible data reuse, and reconciling control over data with desires for open data. In healthcare and nursing, these tensions are particularly pronounced, and existing governance arrangements often prove insufficient to resolve them. However, we still know little about how data-sharing initiatives actually navigate these tensions in practice, and how their governance activities form into overarching orientations, or stances, toward data governance. This study examines four forerunner initiatives that sought to establish data repositories to share data for future AI development. A comparative case analysis reveals twelve concrete data governance practices – situated activities through which stakeholders addressed the specific challenges of sharing data. Patterns across cases revealed that these practices formed three overarching stances toward data governance. We present a progressive, integrative, and protective stance toward data governance, each highlighting specific practices of reconciliation and leading to a variety of outcomes for future AI development. Our findings highlight the importance of aligning normative, organizational, and technical practices and suggest that data governance for sharing data to enable AI development is far from uniform. Instead, it encompasses diverse and situated practices through which responsibility, control, and openness are continually negotiated.
