OpenMDW is a licence designed specifically for AI models, data and weights, with the Linux Foundation seeking formal recognition from the Open Source Initiative.
The Linux Foundation has submitted OpenMDW (Open Model, Data, and Weights) to the Open Source Initiative (OSI) for review and certification. The submission, made in August 2026 by Mike Dolan, senior vice president of legal and strategic programs at the Linux Foundation, seeks formal recognition of OpenMDW as an open-source licence for artificial intelligence systems. The licence itself was released as version 1.1 in May 2026.
OpenMDW was created by the Linux Foundation along with Amazon, Meta, IBM and Microsoft. NVIDIA later adopted the licence across several of its model families, including Cosmos, Isaac GR00T, Ising and Nemotron. Unlike traditional software licences, OpenMDW has been designed specifically for AI systems, addressing components such as model architecture, training datasets and model weights within a single licensing framework.
The need for a dedicated AI licence comes from differences between conventional software and trained AI models. Conventional open-source licences generally cover source code that can be reviewed, modified and redistributed. AI models, however, primarily consist of numerical weights produced through training processes that use large datasets. OpenMDW aims to address these different components through a framework designed specifically for AI rather than applying conventional software licences directly to model weights and datasets.
The OpenMDW submission also raises questions about what qualifies as open source in artificial intelligence. The OSI has previously developed its Open Source AI Definition, but openness in AI remains a debated topic. Issues surrounding the OpenMDW submission include whether redistributing model weights should create obligations similar to those associated with redistributing source code and whether an AI model can be considered open without making its training datasets available.
The OSI had not finalised its decision on the OpenMDW submission when this report was written. The review could therefore become an important step in determining how open-source principles and openness criteria should be applied to AI models, where model architecture, training data and weights play different roles.
