What We Know about Responsible AI Practices in Industry: A Half Decade of Empirical Research

This paper provides a comprehensive synthesis of 161 empirical studies published between 2019 and 2025 regarding how Responsible AI (RAI) is actually practiced within industry settings. By analyzing a wide range of qualitative and quantitative data—including interviews, surveys, and ethnographic studies—the authors map the evolution of RAI from a nascent concern to a professionalized field. The study aims to move beyond high-level ethical principles to understand the ground-level realities of AI development, governance, and deployment.

The authors identify a clear trajectory of progress: RAI awareness among practitioners has grown significantly, and activities such as the use of guidelines, toolkits, and documentation have become more routine. However, this progress is tempered by persistent, systemic challenges. Practitioners frequently report a lack of specialized training and uneven support from their organizations. Furthermore, many existing RAI interventions are perceived as disconnected from the specific demands of their daily work, failing to account for the nuances of different AI pipelines, domains, and product use cases.

The findings suggest that the field must shift its focus from merely cataloging problems to designing RAI interventions as end-to-end sociotechnical systems. The authors argue that researchers should prioritize longitudinal and ethnographic studies to better understand organizational workflows. For policymakers, the study emphasizes the need for governance frameworks that prioritize practical implementability and substantive accountability over procedural compliance, ensuring that regulations are grounded in the actual technical and social realities of industry practice.