A new license drafted by @lf_matt_white from the Linux Foundation has been announced: the OpenMDW License.
This license was specifically crafted for machine‑learning models and their related artifacts. It grants unrestricted, royalty‑free permission to “deal in the Model Materials without restriction, including under all copyright, patent, database, and trade secret rights included or embodied therein.”
This is intended to include activities such as:
Use (run or test) the Model Materials
Copy and modify them
Distribute originals or modifications
The license imposes no field‑of‑use, royalty or geographic restrictions, encouraging broad adoption and innovation.
I wanted to review the OpenMDW license as soon as it was released, but I’ve been quite delayed due to being busy with other matters.
This OpenMDW license is not only a permissive license with aspects very similar to MIT and Apache-2.0, but it also offers the convenience of being able to comprehensively and integrally license AI models, code, and data as a single package. Furthermore, the rights it grants are explicitly stated to include copyright, patent, database, and trade secret rights. I find it a bit strange that it hasn’t yet been submitted to the OSI license approval process, but I believe it is certainly a license that can be called Open Source.
However, it raised several questions for me, so I wrote a rather lengthy review article that includes my perspective on these points:
Is there any legal meaning in including trade secrets within the scope of the grant?
Will there be no side effects from the scope of the patent clause extending to indirect infringement?
Are there no issues with the clause that effectively forces due diligence for the model onto the user?
Isn’t it inferior to Creative Commons when used as a data license?
Doesn’t it have the potential to encourage openwashing?
Personally, I believe all these issues are simply due to a lack of legal testing that comes with OpenMDW’s challenge into a new domain, but I don’t know how other members of the license community will view them. To begin with, I also feel that the OSI community has a limited history of accumulated discussion on intellectual property rights beyond copyright, as well as on personal rights. One could say that the Linux Foundation has given the OSI some homework.
On a final note, I believe OpenMDW will be a useful license framework for cases such as those involving scientific and technical data with patent risks, or where model parameters and data are interdependent to ensure reproducibility. While I expect the existing approach of separated licensing, like MIT/Apache-2.0 + CC, will remain the mainstream, I do consider OpenMDW to be a useful license.