Mozilla and Partners Release Framework to Define Openness in AI Foundations

Mozilla, in collaboration with the Columbia Institute of Global Politics, has published a new framework for open‑source AI in the Communications of the ACM. The paper, co‑authored by more than 40 researchers, builders, and policy experts who first convened in 2024, moves beyond treating AI openness as a single on/off switch. It dissects the AI stack—data, code, model weights, documentation, and more—into distinct layers and introduces a gradient of openness for each. The framework also stresses that safety can only be assessed by examining the entire deployment ecosystem, not just the model weights. By offering a shared vocabulary, it empowers developers, researchers, and regulators to make case‑by‑case trade‑offs without prescribing a universal standard. The authors represent a broad cross‑section of academia, industry, and civil society, underscoring the collaborative effort to make AI systems more accessible, safe, and accountable.