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

[Rob Landley]

On 8/28/26 07:50, Stefano Maffulli wrote:
> I continue getting the impression that the OpenMDW review is tainted by 
> an ideological bias: Because we don't like big tech

The #2 contributor to my toybox project is the Android Base OS 
maintainer at Google. (4223 of my commits, 723 of Elliott's, and 781 
everyone else combined.) I submitted it to SPDX in 2015 because Samsung 
asked me to.

But ad hominem away I guess...

> The question is if OSI thinks that letting every single tiny copyright 
> holder (including the babies) to stop/slow the release of [open weight | 

I'm not sure "you are not big enough, you don't matter" as a position 
you explicitly want OSI to take is really helping your case here.

> Open Source AI | Public AI] (pick whatever term you prefer) is the best 
> way to promote/defend the open source ecosystem.

I think if Disney used a picture of my cat from my personal website in 
their advertising without permission, and I objected, and they responded 
by banning me for life from all disney theme parks citing terms of use 
in some mickey mouse branded bluetooth headphones I'd bought 5 years 
ago, that probably wouldn't make them the good guy.

If you think that's an outlandish example, you missed 
https://www.npr.org/2024/08/14/nx-s1-5074830/disney-wrongful-death-lawsuit-disney

(In case it is not clear to you, I am attempting to highlight the 
potential for abuse.)

> It's been clear to me for years

In a discussion that's like a week old.

> that the view (that you seem to support) 

By all means, ascribe wider viewpoints to me.

> that generative AI is a form of intellectual expropriation will only 
> starve the training data commons, limit ai research, stifle open source 
> AI developers while doing absolutely nothing to hurt, and instead favor 
> the juggernauts OpenAI/Anthropic/Google & co.

AI's need for Lebensraum is irrelevant to whether copyright termination 
for an assertion of copyright infringement is compatible with the open 
source definition.

I would argue that it is not, nor should it be. If this was a term of 
GPLv2, no company would have ever touched it. It attempts to give 
sufficiently popular projects blanket permission to infringe any copyright.

If you want to argue "we didn't REALLY infringe, it was fair use, 
sufficiently clean room, etc" that's great. People have been doing that 
for decades. Often in court. If you want to say "you cannot present your 
case to a judge, you're signing away that right and not just to binding 
arbitration but at all to anybody"... that's not great.

That said, I think it more than 50% likely a judge would grant a 
temporary restraining order against the termination on the theory that 
if they succeed at trial to prove that you DID infringe, then one of 
THEIR remedies may be to stop YOU from distributing the versions 
incorporating the offending material. (Normally a judge just goes for 
monetary damages without taking an existing product off the market, but 
turnabout is fair play. If the license has something like this in it, 
then "if you lose, you have to recall that model and retrain from 
scratch" may be on the table. I'd certainly ask for it, pointing out the 
language in the license as justification. If they think this is 
proportional, then you may owe money AND a recall...)

> The expropriation exists and it's perpetrated by the proprietary models.

And they get multi-billion dollar judgements against them, as the recent 
prominent example I referenced demonstrates.

The little guys generally DON'T get multi-billion dollar judgements 
against them (no point). Deep pockets are most at risk because they're 
worth suing.

If Nvidia just agreed to buy HuggingFace for $12.9 billion, are you 
really defending "the little guy" here? This seems to me an attempt to 
protect deep pockets from those without deep pockets.

(And again, "they're bad so we get to be bad - OSI" isn't an argument I 
would personally be comfortable making in favor of your position.)

> You won't restore a balance without new policies: the answer this time 
> is political, not contractual.

Political as in data center bans...? (Are you coming out in favor of...? 
I'm confused.)

> Back to the license review:

Thank you. That was weird.

> On 8/27/26 16:36, Rob Landley wrote:
>> Under this license, if any of those authors had a day job requiring 
>> them to use Claude, participating in the class action lawsuit would 
>> have either cost them their job or hurt their employer depending how 
>> it was interpreted. 
> 
> Why are you bringing up Claude?

It's the most prominent recent example of a company doing the thing this 
license attempts to enable. They incorporated lots of data sources they 
didn't properly license (which the owners of the copyrights isisted they 
needed to explicitly license), got sued, and paid out.

If this license wasn't created to prevent a recurrence of exactly that, 
what's it for then?

> Anthropic hasn't shared anything about 
> the Claude models. If one of Anthropic models was indeed sharing Model 
> Materials under the OpenMDW then I know I wouldn't be on the side of the 
> large copyright holders (the same way that I was on the barricades 
> against RIA/MPAA during the Napster wars that brought us the "lovely" 
> freedom-depriving DMCA).

Ah, citing the deep magic. I'm old too.

https://www.fool.com/archive/portfolios/rulemaker/1999/03/18/the-rule-maker-portfolio.aspx

> BTW, Anthropic solved the issue by buying the 
> books, the issue is solved the same way that Google solved it with Books 
> years ago.

I'm mostly subscribed here in case OSI misnames 0BSD for a third time. 
Google, Microsoft, Facebook, Intel, and so on have released projects 
licensed under 0BSD. It was never just "for me" or "about me". I was 
aware how people who think VERY DIFFERENTLY than me could use or misuse 
it. (The #1 ranked project in 0BSD on github is written in _rust_. Sigh. 
Oh well.)

If this OpenMDW license becomes OSI approved, do you believe future 
models from Antropic and OpenAI won't include similar termination 
language in their models' licenses, and next time they're in court point 
to the Open Source Initiative's endorsement of copyright termination in 
response to copyright enforcement as a reason to uphold its 
enforceability? (When arguing why the judge _shouldn't_ issue a 
preliminary injunction putting said termination on hold pending the 
outcome of the litigation.)

> These examples ad absurdum don't help understand the limits of the license.

One example, from this year, for billion dollars, is ad absurdum?

My sense of time is a bit off since the pandemic, but I believe this 
license was submitted for discussion to OSI _right_ after the anthropic 
settlement?

https://www.npr.org/2026/07/27/nx-s1-5904606/anthropic-vs-bartz-ai-copyright-lawsuit-pros-cons

Yeah, July vs August. One month. Are you saying this license is not in 
any way motivated by the anthropic settlement?

> /stef

Rob