Mozilla and the Columbia Institute of Global Politics have published a peer-reviewed framework in Communications of the ACM titled 'Unpacking Open Source AI: Toward a Framework for Openness in Foundation Models.' Building on a 2024 convening of 40+ researchers, builders, and policymakers, the paper argues that AI openness should be treated as a gradient across components (data, code, weights, documentation) rather than a binary label. It also argues safety must be assessed at the system level, including deployment environment and governance, not just model weights. The framework avoids prescribing a single correct level of openness, instead offering shared vocabulary for developers, researchers, and regulators to evaluate tradeoffs case by case.
