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

John Colagioia jcolag at disroot.org
Sat Aug 29 14:34:09 UTC 2026


As a newcomer, and on the digest-track, for (primarily) the OpenMDW License issue, I apologize if I cover any ground that others have discussed or otherwise transgress.

  

In any case, as I read the license (non-lawyer, but capable of reading carefully), I see several fundamental problems with calling OpenMDW an Open Source license.  Most critically?  I don't see any evidence that an AI model can qualify for copyright.  Models consist mainly of tensors describing transformations of vectors, in other words a list of mathematical formulas, expressly forbidden from receiving copyright, at least in the United States.  And they're not made by humans, so wouldn't qualify for copyright protection, even if you could get a "pile of formulas" copyright.  Without a copyright, then what rights does the license even grant?

  

A corollary to the lack of human input, models have no reasonable "source code" to open, removing even the possibility of a license offering Freedoms 1 and 3 in any reasonable sense.  A developer could plausibly patch a GGUF-ish file, but that seems like a pathological kind of "open."

  

Ignoring that, the license admits twice that the vendor may not even have the right to distribute the model.  The anti-lawsuit clause (¶5) and the second disclaimer (¶8) both assert in different ways that the vendor refuses to take responsibility for the copyright status of the work itself, passing responsibility downstream.  No significant public license asserts that the vendor has the right to distribute the product, but none raises the question so deliberately.  And this goes back to the initial problem:  If the vendor explicitly refuses to acknowledge that they own the model and have the right to distribute it, then what rights can the license grant?

  

Slight tangent:  17 USC §506(c) says that "any person who, with fraudulent intent, places on any article a notice of copyright or words of the same purport that such person knows to be false...shall be fined not more than $2,500."  I don't think that anybody wants to defraud anybody with the license, but the implied claim of ownership of the model in offering the license while also asserting the right to not own it feels like it sets up every vendor for exactly that lawsuit.

  

Those two paragraphs also not-coincidentally serve to discriminate against users.  ¶5 says that the vendor revokes the license if the user participates in a lawsuit against the vendor.  ¶8 says that the user can't use the model for illegal purposes.  The latter seems especially odd, since a license that did try to grant permission for illegal actions would be invalidated.

  

And even if we consider the anti-lawsuit thing legitimately "open" after decades to the contrary, it seems far too broad without a strict definition of "participate."  Witnesses, jurors, and the entire defense participate in lawsuits, but could all have their work impacted if their employer requires use of an LLM released under the OpenMDW's terms and involved in their case.  "They don't mean that," sure, but not providing bright-line definitions means that the vendor has oversight over who uses the product, in other words NOT a public license.

  

Overall, I don't see how Open MDW works as an open source license.  The model probably doesn't qualify for a copyright.  If it does, then the license suggests that the vendor probably doesn't own it, opening the vendor up to fraud accusations.  If it qualifies for copyright and the vendor can justifiably claim to own it, then it has no source material to sensibly make it "open source."  Likewise, if it excludes people from taking advantage of the license for any reason, then it can't sensibly call itself "open source."  Or rather, if it fits the criteria as an open source license, then I can't think of a single public license that would fail to meet those criteria.

  

Thanks!
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opensource.org/pipermail/license-review_lists.opensource.org/attachments/20260829/de549b78/attachment.htm>


More information about the License-review mailing list