Unpacking Open Source Artificial Intelligence: Toward a Framework for Openness in Foundation Models

There has been a robust debate about the benefits and risks of “open sourcing” foundation models. Yet much of the discourse has lacked specificity, either remaining overly abstract or focusing narrowly on a limited set of technical components. This is partly because defining open source for foundation models is challenging, given their significant differences from traditional software development. To inform a more practical and nuanced discussion of openness in AI systems, including foundation models, we propose a framework for analyzing openness across the AI stack. We review prior approaches and highlights the diverse motivations for pursuing openness, showing how these map onto different forms of openness at both the model and system levels. The framework makes these distinctions explicit and provides a foundation for evaluating how openness strategies advance broader goals such as safety and accountability.