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

Thu Sep 3 20:55:58 UTC 2026

Sado-san, Yokoi-san, thank you both for your detailed questions about the
scope of the termination trigger.

1. "Directly or indirectly infringe" (Sado-san).

Your first reading is the intended one. "Indirectly" covers secondary
liability claims: for example, a claim that making, using, or distributing
the Model Materials helps or causes other people to infringe someone's
rights. You are right that, technically, this kind of liability belongs to
a person rather than to materials. Apache-2.0 uses the same shorthand. Its
Section 3 also references “the Work” itself (rather than an entity’s use,
distribution, etc. of the Work) as something that "constitutes direct or
contributory patent infringement," and we followed the same wording style.

Other common FOSS licenses tend to use comparable language referring to the
licensed thing itself when speaking of termination for infringement, e.g.
EPL-2.0 section 7 (“...alleging that the Program itself… infringes such
Recipient’s patent(s)”), CDDL-1.1 section 6.2 (“...alleging that the
Participant Software… directly or indirectly infringes any patent”),
MPL-2.0 section 5.2 (“...alleging that a Contributor Version directly or
indirectly infringes any patent”).

The OpenMDW-1.1 licenses the Model Materials only, and the termination
provision requires a claim against the Model Materials. OpenMDW-1.1 does
not reach back to artifacts used in the creation of the model unless such
artifacts are included in the subsequently published Model Materials that
are themselves licensed under OpenMDW-1.1.

2. Re: the termination boundary (Yokoi-san).

Taking your questions in order:

1) Correct. A lawsuit claiming infringement of a database right does not
trigger termination, and the license grants stay in place. The trigger
covers only claims that the Model Materials "infringe any patent or
copyright." Database rights are a right available in the EU, the UK, and a
small number of other jurisdictions, but they are far from universal, and
the US has no equivalent. We looked at this closely when we drafted the
CDLA data license agreements.

2) Correct. A trade secret lawsuit does not trigger termination. At least
in the US, a trade secret claim is "misappropriation," not "infringement,"
so it is outside the trigger. There is also a practical reason to leave it
out as publishing materials openly destroys secrecy. A trade secret claim
about openly published model materials makes little sense as a realistic
threat.

3) Your first two questions lead directly to the principle. The trigger is
not designed to match the grant. It was designed to match the real
litigation threats that openly published model materials face. Trade secret
claims are not relevant because of the secrecy point above. Database rights
are an odd situation because a license used worldwide should not make
termination depend on a right most countries do not offer.

The realistic lawsuit against model materials is still a copyright lawsuit
or a patent claim against the functionality.

I'd also point out that Apache-2.0 works on the same principle. It grants
both copyright and patent licenses, but its termination trigger covers only
patent claims. As discussed earlier in this thread, the Apache-2.0 trigger
followed the prominent threat of its time, not the scope of its grant.

I recognize the debate over OpenMDW-1.1 has extended into a broader policy
on termination. My personal position is that adding copyright to the
termination trigger does not imply that other claim types should follow.
Copyright was added because that is where the very real threat to model
publishers exists. Patent claim threats to model publishers are likewise
real. We discussed this at one point in drafting, and this is where we
ended up.

Mike

---
Mike Dolan
The Linux Foundation
Cell: +1.440.552.5322
mdolan at linuxfoundation.org

For help scheduling a meeting, please contact Jackline Mbithi <
jmbithi at linuxfoundation.org>.

---

On Wed, Sep 2, 2026 at 12:18 PM Shuji Sado <shujisado at gmail.com> wrote:

>
> Mike-san, one point of clarification on your summary in 7(b). I realized
> that I may have been reading one part of Section 5 too narrowly. When the
> license says that the Model Materials “directly or indirectly infringe”
> copyright, what exactly is intended by “indirectly”?
>
> In copyright law, indirect infringement ordinarily makes me think of
> secondary liability such as contributory or vicarious infringement, which
> is generally liability of a person or entity rather than something that the
> material itself “does.” Is “indirectly” intended to cover a claim that use
> or distribution of the Model Materials induces or contributes to
> third-party infringement, or is it intended more broadly, for example to
> reach claims based on infringement occurring during the training process?
>
> Shuji
>
>
> 2026/9/1 23:00 Michael Dolan <mdolan at linuxfoundation.org>:
>
>> A few points raised since Friday were new, so I will share my view on
>> those in case it helps everyone's understanding. It’s challenging to
>> keep up with all the points raised and comments. I will then summarize our
>> position regarding the more contentious points and areas of confusion that
>> resurface throughout this thread.
>>
>> 1. "Voluntarily participate" (Richard's point).
>>
>> Richard, you are correct in tracing this to the patent non-assert
>> language from the Microsoft promises through OWFa 1.0, whose agreements we
>> use today across a number of specification communities. For example, the
>> Open Container Initiative releases specifications under the OWFa 1.0
>> patent-only variant alongside Apache-2.0, and GraphQL is licensed under
>> OWFa. The phrase does specific, narrow work. "File or maintain" alone would
>> invite an obvious dodge: fund, direct, join, or control litigation
>> nominally brought by someone else. The word "voluntarily" excludes
>> compelled participation. A subpoenaed witness, a deposed employee, a juror,
>> and a party joined involuntarily do not voluntarily participate. I can only
>> confirm our intent in OpenMDW. David Rudin, who chimed in earlier on this
>> thread, was part of drafting the OWFa (and maybe also the Microsoft
>> promises). I'll defer to him on the OWFa's intent - we weren’t involved in
>> that drafting work.
>>
>> 2. "Corresponding lawsuit" (Richard and John).
>>
>> The termination provision targets voluntary first aggression, so a party
>> who was sued first does not lose its license by responding, whether by
>> counterclaim, cross-claim, or a separately filed responsive action.
>> "Lawsuit" rather than "counterclaim" is deliberate. Responsive claims
>> sometimes must be brought as separate actions, in a different forum or a
>> declaratory posture, and the exception should not turn on civil procedure.
>> "Corresponding" ties the responsive suit to the litigation first brought
>> against you. Without it, any unrelated suit anyone had ever filed against a
>> party would immunize that party's later aggression indefinitely. I would
>> also note that compulsory counterclaim rules in some jurisdictions require
>> a defendant to assert related claims or lose them. The exception ensures
>> the license never forces a choice between complying with those rules and
>> keeping the license. So Richard is right that it does not require the two
>> suits to share underlying subject matter, and it is not merely "another":
>> it means responsive to the action first brought against you.
>>
>> 3. Use in litigation. (Kevin’s points)
>>
>> Kevin, regarding your concern about retained experts and counsel and
>> access in litigation, I had a couple of reactions. First, termination
>> operates only as of filing or participation, so all pre-suit investigation
>> and testing happens under a full license. Second, use as evidence after
>> filing does not depend on the license.
>>
>> Use for purposes of judicial proceedings is protected independent of any
>> license, in the US as fair use, and analysis of accused materials proceeds
>> through discovery. This isn’t just a license steward’s claim; see, e.g.,
>> Bond v. Blum, 317 F.3d 385, 392-397 (4th Cir. 2003) (finding fair use of
>> copyrighted materials when used as evidence in a court proceeding), and I’m
>> sure dozens of other cases out there.
>>
>> I’ll point out that the same situation would apply, for example, for a
>> case relating to software under MPL-2.0 where a plaintiff brings both
>> patent and copyright claims (the Oracle v. Google litigation is an example
>> where both types of claims were brought together, albeit not for
>> MPL-2.0-licensed software). Section 5.2 of MPL-2.0 would terminate all
>> license grants (both patent and copyright) due to the patent claim. The
>> plaintiff would presumably have to investigate their copyright claims using
>> mechanisms other than an explicit license grant under MPL-2.0. I can't
>> claim to know how that would work outside the U.S., but at least in the
>> U.S., we have other avenues than the license grant.
>>
>> 4. The API and RAG hypo (Moming's point).
>>
>> Moming, I don't think your hypothetical works as stated under the
>> definition of Model Materials. Model Materials means the materials
>> "provided to you" under the agreement. A provider's server-side store of
>> user inputs is never provided to the user and is therefore not part of any
>> user's Model Materials. A user of a hosted API typically receives no Model
>> Materials at all and needs no copyright license to use a hosted service;
>> the provider's terms of service govern that relationship, not this license.
>> A claim that a provider is misusing your inputs is a claim about the
>> provider's conduct, not a claim that materials provided to you infringe. I
>> do not see how OpenMDW would apply in this scenario, as that conduct likely
>> falls under their terms of use or service agreement.
>>
>> 5. Downstream recipients (Carlo, in response to your point (a)).
>>
>> Yes, the definition and grant are per recipient. Each recipient's license
>> runs directly from the providers under that recipient's own agreement, not
>> through a chain from an intermediate distributor, and termination reaches
>> only the rights and grants made to the litigant. A downstream recipient who
>> is not a participant in the litigation holds their own grant and is
>> unaffected.
>>
>> I would note again that OpenMDW licenses are already in use in the field,
>> and many licensors have published model materials under them. I accept your
>> point that it could be clearer, but I'm also not aware of any perfectly
>> drafted open source license… so I'll request a similarly fair evaluation as
>> other licenses in use, warts and all. :-)
>>
>> 6. On the software freedom subthread, I'll stay out of the definitional
>> debate, but Luis, your reframing toward the health of the software commons
>> resonates with me. That is the test this license was built for: moving
>> model publishers from bespoke restrictive terms into the commons. I
>> couldn’t justify to a business decision-maker that they couldn’t use
>> OpenMDW-1.1 licensed models because of the arguments playing out in this
>> thread. At some point in the analysis, this thread has gone to a strange
>> place.
>>
>> 7. Finally, since several earlier points keep resurfacing as the thread grows,
>> and it appears people are not going back to read everything (which is
>> understandable), I thought I’d provide a short summary of our positions
>> as the steward, which we have already stated in this thread. Perhaps this
>> will help bring this debate to a close:
>>
>> (a) The termination provision does not preclude or waive any claim or
>> remedy, and it is merits-neutral. It presumes nothing about whether any
>> claim, patent or copyright, is valid, and it operates identically either
>> way, as patent retaliation always has. It does not prevent anyone from
>> filing any litigation on any topic. It ends only the claimant's own license
>> to the materials claimed to be infringing upon filing such a lawsuit.
>> Pre-litigation enforcement is also out of scope. The license termination
>> provision puts the parties back into the same position they were in before
>> any license was conveyed.
>>
>> (b) The trigger reaches only claims that the licensed materials
>> themselves infringe, the class of claims that attacks the subject of the
>> grant itself. Broader triggers, such as any IP claim or any lawsuit, were
>> considered and rejected.
>>
>>
>> (c) Terminating copyright grants as a remedy appears in OSI-approved
>> licenses (e.g., MPL 1.1 §8.2, CDDL 1.1 §6.2, OSL 3.0 §10, CAL-1.0 §5.3),
>> and OSI has approved a trigger broader than patent claims in OCLC-2.0 §5.
>>
>> (d) OpenMDW does not relicense third-party code, and it doesn’t “grab”
>> third-party code into its scope. The grant conveys only rights the
>> providers hold or can pass through for the materials identified as being
>> provided, and third-party components keep their own licenses and notices.
>> Unlike other licenses, which may have narrower grants, the grant is
>> intended to encompass all possible rights a user of an OpenMDW-license
>> model may need to practice/use the model. This is intentional and designed
>> to overcome potential lack of clarity in other licenses when used in an AI
>> model context. For example, Apache-2.0’s grants are centered on copyright
>> via “the Work” which is “the work of authorship” (copyright), and then the
>> patent grant is scoped to “the Work”.
>>
>> (e) “Model Materials” is transaction-scoped to the materials provided to
>> you under one agreement. Separate releases are separate agreements, and
>> there is no ever-expanding covered work. The definition's own word
>> "related" bounds the unit of materials concern that keeps coming up. I
>> understand that the OpenMDW is novel in the scope of Model Materials, but
>> honestly, for how many years have we debated “what is the Work” under
>> Apache-2.0 or “the Program” under EPL-1.0 without ever questioning if they
>> satisfy the OSD requirements?
>>
>>
>> (f) The clearance paragraph is a risk-allocation disclaimer. The license
>> has one affirmative condition, notice retention, and one termination
>> mechanism, the litigation provision. Where this license family intends a
>> condition, it says so expressly. And licenses such as §2(c) of EPL-1.0 and
>> EPL-2.0 show that even if such risk-allocation terms were an explicit
>> condition of the license grant (which they are not for OpenMDW), an
>> explicit condition would still not conflict with the OSD.
>>
>> (g) We can clarify many of the concerns raised here with FAQ updates
>> stating these interpretations and publishing guidance for model publishers
>> that recommends enumeration of each release's Model Materials.
>>
>>
>> ---
>> Mike Dolan
>> The Linux Foundation
>> Cell: +1.440.552.5322 <(440)%20552-5322>
>> mdolan at linuxfoundation.org
>>
>> For help scheduling a meeting, please contact Jackline Mbithi <
>> jmbithi at linuxfoundation.org>.
>>
>> ---
>>
>>
>> On Tue, Sep 1, 2026 at 1:19 AM Luis Villa <luis at lu.is> wrote:
>>
>>> On Sun, Aug 30, 2026 at 2:51 PM Pamela Chestek <pamela at chesteklegal.com>
>>> wrote:
>>>
>>>> McCoy, you asked for a definition of "software freedom" and Moming's
>>>> response is similar to my thoughts:
>>>>
>>>> On 8/29/2026 11:10 PM, Moming Duan wrote:
>>>>
>>>> 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?*
>>>>
>>>> My premise is that the everyone benefits from a software commons, so,
>>>> does the license encourage or discourage sharing?
>>>>
>>> "Software freedom" is the ability to do anything you want with the
>>>> software. A few conditions/impairments are allowed because they indirectly
>>>> encourage the growth of the software commons - giving credit (copyright
>>>> notice/attribution), telling people what their rights are (the license),
>>>> the appropriate distribution of risk (the "as-is" rep and disclaimer of
>>>> warranty), and compulsory sharing (copyleft). The balance is very delicate.
>>>> It's questionable whether the GPLv3, and then the AGPL, contribute to the
>>>> health of the open source software commons because they are commonly used
>>>> as the antagonist to encourage people to get a commercial license instead.
>>>> It does, though, seem that lever isn't as successful anymore, since people
>>>> are more willing to comply with the copyleft licenses and these two
>>>> licenses are more often being used for their intended benefit.
>>>>
>>>> This, I think, is the piece that's missing from the OSD. We've seen the
>>>> OSD gamed when it has been formally met but it is nevertheless clear that
>>>> the license is designed to give one entity superior rights in the software
>>>> or has mechanisms that will discourage the use of the software. "Providing
>>>> software freedom" as a requirement is a check on that.
>>>>
>>> FWIW any or all of Pam's formulations of "encouraging sharing of
>>> software" or "contributing to the health of the software commons" or
>>> "protecting the open software ecosystem" (from an opensource.com essay
>>> of yours, Pam, IIRC) would be vastly superior to "providing software
>>> freedom".
>>>
>>> Besides the obvious point that software doesn't have freedom, any of
>>> those would be much clearer to argue about, with less baggage. (There would
>>> still be plenty of arguing; all of those still require value judgments. But
>>> at least it'd be a value that one can attempt to decipher from the words.)
>>> _______________________________________________
>>> The opinions expressed in this email are those of the sender and not
>>> necessarily those of the Open Source Initiative. Communication from the
>>> Open Source Initiative will be sent from an opensource.org email
>>> address.
>>>
>>> License-review mailing list
>>> License-review at lists.opensource.org
>>>
>>> http://lists.opensource.org/mailman/listinfo/license-review_lists.opensource.org
>>>
>> _______________________________________________
>> The opinions expressed in this email are those of the sender and not
>> necessarily those of the Open Source Initiative. Communication from the
>> Open Source Initiative will be sent from an opensource.org email address.
>>
>> License-review mailing list
>> License-review at lists.opensource.org
>>
>> http://lists.opensource.org/mailman/listinfo/license-review_lists.opensource.org
>>
>
>
> --
> Shuji Sado
> Chairman, Open Source Group Japan
> https://opensource.jp/
> English blog: https://shujisado.org/
> Japanese blog: https://shujisado.com/
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opensource.org/pipermail/license-review_lists.opensource.org/attachments/20260903/0491655c/attachment-0001.htm>