[License-review] For Approval: OpenMDW License Agreement, versions 1.1 (OpenMDW-1.1)

[Moming Duan]

Hi all,

Since MG came up in this thread (McCoy's and Richard's recent messages), let me offer a few quick clarifications. I'd rather not split the thread while the focus is on OpenMDW, so if you have questions about ModelGo itself, please join us in the MG thread.

  1.  Yes, the "Licensed Materials" definition includes Complementary Materials, but they are there to support the Model. The license is meant for publishing models, as I clarified in the MG thread.
  2.  MG0 has no termination clause. MG-BY has one, but it is triggered by patent litigation only, not copyright. That discussion already happened in the MG thread.
  3.  I have withdrawn MG-BY-SA from review. After discussing it here I agree this question is premature: there is no reliable way to verify whether distillation happened, so copyleft for distilled models is hard to make workable.

________________________________

Now my view on OpenMDW-1.1.

First, I am glad to see OSI and the community taking AI model licensing seriously. I hope we will see more genuinely Open Source model licenses. Many proprietary model licenses put "Open" in their names, and that confuses people. I keep a watchlist of those in my LicenseAtlas project; the Databricks Open Model License is one example: https://morningd.github.io/license.atlas/licenses/databricks-open-model-license

As for the status quo: models released under MIT or Apache-2.0 (e.g., GLM, DeepSeek) are already quite satisfying to me, although I think MG and OpenMDW-1.0 are better written in some respects. (I stopped using GPT, after upgrading to 5.6, its guardrails refuse to help with the tedious form-filling in my university's online systems.)

I use AI tools every day in my research. A permissive license matters a lot to me: it means I can trust the model and build my AI agents on it (recently an expense-reimbursement assistant) without worrying about legal disputes. I think this is also why some people in this thread feel so strongly.

My concern with OpenMDW-1.1 is of course the copyright-termination clause. It is serious enough that I would avoid models under it. As I read the license (I am not a lawyer), simply using the model means accepting it, so if I access the model through an API key, I am a licensee. Now consider a scenario:

A cloud provider deploys the model and I connect through an API. During use I find out that my inputs are being used to improve the model. Not necessarily post-training; storing my inputs in their database for RAG counts too. Because I accepted the license, I cannot pursue a copyright claim.

By then I may have deployed the model widely inside my organization, and switching is never free. See my GPT example. Some businesses even run quantitative trading on models.

It could get messier: imagine I run a patent agency, and the cloud provider takes my clients' patent content.

I am imagining things, and I am not a lawyer, but as a developer this is how I would read the clause, or misread it. Either way, I would not trust models published under this license. It takes from me before I even figure out what it provides.

________________________________

Here is a simple test of my own: if 99% of the world's models were released under license X, would our ecosystem be better off?

  1.  MIT: fine. Even if proprietary software free-rides on it.
  2.  Apache-2.0 or GPL-3.0: also fine. Most people hold no patents.
  3.
OpenMDW-1.1: my options would be to give up asserting my copyright, give up AI, or hope that AI companies are kind.
Would that be a better Open Source ecosystem?

You may find this extreme. I think that is exactly why this thread got so heated: some people look further ahead, some look at the present. My personal view is that for now, we need a good starting point, and approving one more open license hurts nothing. That is also why I stripped most clauses out of ModelGo and made MG0.

Best,
Moming Duan

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opensource.org/pipermail/license-review_lists.opensource.org/attachments/20260830/69a26ca7/attachment.htm>