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

[Richard Fontana]

On Thu, Sep 10, 2026 at 9:09 PM Pamela Chestek <pamela at chesteklegal.com> wrote:
>
> I do find it inconsistent with my understanding of open source and the open source community that someone who intentionally committed a wrongful act should be allowed to avoid the consequences of that act. I raised this point early on and no one responded to give a reason why this intentional conduct was innocent enough that the withdrawal of the license, in Joshua's words, is a price that a licensor can demand in exchange for the use of their work. We expect members of the open source community to act ethically, but this provision allows deliberate, intentionally, willful wrongdoers to escape some degree of responsibility for their choices merely by choosing this license. This is not an AI vs. software issue, this is true if the license is used for software alone, as it can be. I don't understand why that is acceptable in a license and would very much like to understand from others why they think it is.

Thinking about Luis's comment about Oracle/Google: I would assert that
in FOSS, copying and reimplementing APIs has, in a non-strictly-legal
sense, basically not been considered "wrongful" by those carefully or
intuitively considering the question, going back to the dawn of time,
at least prior to Oracle reviving the theory of API copyrightability
(Oracle, arguably unlike Sun in this specific context, was part of the
FOSS community). So if a purportedly FOSS license had a provision that
said "if you try to sue me via a copyright claim for reimplementing
what you contend are 'your' APIs, you lose all the copyright licenses
I'm granting you", maybe that would be okay. OpenMDW-1.1 is different
in this respect because its copyright-triggered termination reaches
*all* sorts of copyright claims, and I recognize the point Mike made
that it would probably not be feasible to have a more narrowly
tailored one that distinguished different types of claims.

The justification given for (what I continue to contend is) the
unprecedented copyright-triggered termination in OpenMDW-1.1 is
centered around the current risk situation faced by model publishers.
Let's be clear that what this means is publishers of pretrained model
weights (and probably in practice only those sorts of models that are
trained on a vast amount of data). The other copyrightable materials
released by model publishers, in connection with their release of
model weights, do not, of course, face this situation. No one seemed
to think much of my highlighting of that Nemotron model latex2html
Python script, but am going to stubbornly insist that it's a good
example of why this license is problematic. The asserted policy
justification for copyright-triggered termination has no relevance to
latex2html.py, and yet it is swept in improperly by this termination
provision. It's true that if I care a lot about this issue, I can try
to go to NVIDIA and get them to release latex2html.py under the MIT
license, but that is not likely to be successful, in part because it
would defeat the theoretical purpose in selecting this all-purpose
license to begin with.

Maybe the OpenMDW-1.1 meets the expectations the OSI had for terms for
*parameters* in the OSAID: "The Open Source AI Definition does not
require a specific legal mechanism for assuring that the model
parameters are freely available to all. They may be free by their
nature or a license or other legal instrument may be required to
ensure their freedom. We expect this will become clearer over time,
once the legal system has had more opportunity to address Open Source
AI systems." (OSAID-1.0) See also:
https://opensource.org/ai/checklist, where it is quite clear that the
OSI contemplates that things other than OSI-approved licenses might be
suitable for model parameters, but that does not apply to code or
model architecture materials.

I believe it was Sado-san who pointed out that the OSI had approved
the SIL Open Font License, as an example of a license that is designed
and contemplated mainly for a special type of material that is not
software in the conventional sense, but I would assert that this
license (while possibly acceptable in the scheme of things for
quasi-open fonts, given the sad history surrounding that topic) was
wrongly approved by the OSI, because of its "Sun RPC" clause ("Neither
the Font Software nor any of its individual components, in Original or
Modified Versions, may be sold by itself"). I apologize 17 years later
for having played a small undocumented part in that approval, in the
form of encouragement of Michael Tiemann to promote the license for
OSI approval. The OSI does not currently have a process for de-listing
licenses, and I wouldn't put OFL-1.1 at the top of the list if it did
because there are actually worse licenses on the OSI-approved list,
but it would be on the list.

However, for better or worse, the OSAID is completely irrelevant here.
The license steward is not submitting this license as a potentially
OSAID-conformant license aimed solely at model parameters, but as a
normal potential OSI-approved license, that can apply to software
having no connection whatsoever to AI models, or indeed other kinds of
copyrightable materials that are commonly placed under open source
licenses. The prospect of OpenMDW-1.1 being used as a license for some
or all parts of an OSAID-1.0-conformant model release is, I assume,
extremely remote.

The OSI must consider how a submitted license will operate when
applied to ordinary software. OpenMDW-1.1 is flawed in its basic
conception, as a consequence of the termination provision. Its whole
raison d'être is to have a single license that applies to everything.
But IMO that suggests that the policy justification for the
termination provision is not persuasive because that policy
justification only truly applies to model weights - leaving aside
whether the justification has merit in that specific context.

 Richard