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

[Josh Berkus]

On 8/26/26 4:59 PM, Pamela Chestek wrote:
>>
>> 2. On "someone whose code was infringed cannot bring a claim": that is 
>> not what the provision does, and I want to be precise about it because 
>> the distinction matters. B can bring the claim, including for willful 
>> infringement, with every remedy intact, and A must defend it fully. 
>> There is no immunity of any kind. What B cannot do is keep exercising 
>> A's license to X while maintaining the suit.
>>
>> Practically, I have to look at the consequence of that choice for a 
>> genuinely wronged developer. A developer who sues A asserting that X 
>> was built by deliberately and unlawfully copying the developer's code 
>> is asserting that X is itself unlawful. At the same time, many 
>> alternative models are available to the developer – X is not even 
>> remotely their only option. A developer with a genuine copyright claim 
>> has no realistic need (and I would suspect… no desire) to keep 
>> building on the very model they have told a court is the product of 
>> copyright infringement. For that developer, the provision costs 
>> nothing they actually want – why would they want to use or distribute 
>> a model someone else created and published that infringes their copyright?

> This may be the crux of the problem that I have (and some developers on 
> social media and on this list have). I agree and apologize for 
> mischaracterizing the legal effect, and thanks for calling me out on it 
> (not for the first time either!). But I believe that "but there are 
> other options!" is a red herring. I was thinking of the theoretical 
> situation where the license is widely adopted. Assume for the sake of 
> argument that all AI models have either proprietary licenses or this 
> license. This puts the FOSS developer in the situation of either not 
> using AI at all or being forced to use a proprietary model if they still 
> want to have a claim against the open source model. I believe the fact 
> that this license drives developers to a proprietary solution rather 
> than an open source one is an indication that there is a systematic 
> failure of the license to support the open source ecosystem.

I got 99 problems, but a copyright antisuit clause ain't one.

Before LLMs and the Anthropic suit, a copyright antisuit provision would 
not have been controversial.  If someone had added a clause identical to 
this to a license submitted 10 years ago, this forum would probably have 
approved it without much argument as a simple extension of the patent 
grant antisuit provision.  But, here we are.

The thing is, the antisuit provision doesn't apply to all OpenMDW 
models, it applies to the *specific model whose distributor you're 
suing*.  So say this became the most popular AI and model license, and 
there are 100 different models under the license.   If I sue the creator 
of model #37 for copyright infringement, I lose my legal right to use 
model #37 ... but not the 99 other models, even if some of those 99 
other models belong to the same organization (at least as I read the 
wording).

To rephrase: as I read it, you don't lose access to all OpenMDW models 
if you sue over just one of them.  Just the one you sued.

As a developer who could be in either seat -- the downstream developer 
contemplating a lawsuit or the model creator possibly being sued -- this 
doesn't seem controversial to me.  It seems pretty fair, actually.  If I 
was the open source developer of a new model, I would not want someone 
facing me in a copyright lawsuit to be able to use the model I created, 
for free, while suing me.

So, thought experiment, when does this become a problem?

Well, there's the hypothetical of "the Linux of models", where a single 
OSS model becomes so industry-dominant that inability to use it is a 
serious handicap to someone's ability to find employment. While that 
doesn't seem likely right now, neither did Linux in 1995.  However, that 
isn't different from a patent retaliation clause -- there was a time in 
the industry (roughly 2003) when suing Apache would have made it 
impossible to have a web business.  And we've approved lots of patent 
antisuit clauses.

Since we are talking about AI, this does create the weird quirk that an 
OpenMDW legal help model cannot be used to help prepare a lawsuit 
against its creator, which is a restriction on use.  But, again, not 
different from patent clauses, and what seems like a reasonable carve-out.

-- 
Josh Berkus