@EUCommission

The #EuropeanCommission advertises as "open-source" an #LLM which -by their own call- need only provide, "where applicable, the release of open weights, .. OR .. technical documentation" [1].

So the result is open-weights at most, and would not even satisfy the OSI Open Source AI Definition (#OSAID 1.0).

This is not the first time the EC's claims don't survive scrutiny.

An institution funding "sovereign AI" at scale should know the difference.

[1] https://aiboost-project.eu/wp-content/uploads/2026/02/AI-BOOST-_Guidelines-for-Applicants_FRONTIER-AI-CHALLENGE.pdf

Some might ask why we are accepting something that deviates from the #OSAID.

We are doing so for the same reason we made the #OSAID, despite knowing the first draft would likely not be perfect: because something is needed now, and this is much better than the status quo (government saying nothing).

Even if this definition, (or even the OSAID) aren't perfect, they fundamentally strengthen protection against Open Washing, and enhance the four freedoms. And that is our mission 🫡.

When it came to Open Source AI, the #OSI's #OSAID was the inspiration for the work. That means models’ weights,
deployment code, training code, and at least data information (but preferably training data).

I defended the #OSI view that Data Information is a necessary compromise for cases where data sharing is impossible for technical or legal reasons (copyright, data protection etc..), to ensure legal certainty.

But also shared that a part of the community has reservations about the approach⬇️

Open Source Initiative brings its expertise on openness in AI to the G7
https://opensource.org/blog/open-source-initiative-helps-g7-deliver-vision-on-ai-openness #AI #OSI #OSAID
Open Source Initiative Helps G7 Deliver Vision On AI Openness

On May 29, 2026 in Paris, G7 Digital and Technology ministers approved a “Vision on AI openness opportunities and shared language”. The vision, which sets out terminology around AI openness, is the…

Open Source Initiative
In advance of the first in person meeting in Paris, we immediately got to work supporting the French Presidency with drafting, but were also concious that while the #OSAID had garnered support, it was not unanimous, and so it was important to bring the voices of all #OpenSource communities to the table. ⬇️
#OSI literally maintain a list of "approved" #opensource licenses.

They are very corporate friendly. More precisely, #BigTech friendly.
In fact, they refused #SSPL when #Amazon was among their top sponsors and adopted the much more subversive #CAL when #HoloChain was among their top sponsors.

Furthermore, I partecipated to the discussions around the #OSAID some years ago and I've never met such a corrupt environment before. They used absurd tricks to please #Google and #Meta and not include "training data" among the requirements, so that they could easily break free of #AIAct.

So, they are not just gatekeepers, but basically unregistered BigTech lobbyists.

CC: @[email protected]

@tante The "Open Source AI Definition" was the final straw for me to recognize OSI for what they are (at least have become): a capitalist apologist shill group.

I do not like that this is true.

https://opensourcerer.eu/osaid-v1-0-notes/

#OpenSource #OSAID #OSI

OSI's Open Source AI Definition v1.0 | OpenSourcerer.eu

Lars Marowsky-Brée on Free, Libre, and Open Source, Software Engineering, and IT

@[email protected]

First it's important to note that over years I realized that #OSI is just a corporate (and US-led) gatekeeper organization that serve the very interests their sponsors.
You can easily see this reading their license review mailing while keeping a tab opened on the sponsors page of the day through the #WaybackMachine.
Just as a couple of example, they rejected #MongoDB's #SSPL while #Amazon was their major sponsor and adopted CAL that was way more contentious.

The last damage that OSI did to our communities has been the #Meta dictated #OSAID (OpenSource #AI Definition) better known as #OpenWashing Definition, that superseed the #OSD and does not require training data sharing, voiding the freedom to study and welcoming toxic candies within "open source" just to avoid the #AIAct requirements.

So I don't care about OSI opinion about the #HackingLicense (or about anything else).

Having said that, you are right that its first condition forbid any use of the covered work that would limit third party access or use of it.

So basically you can't use your freedom to limit the freedom of others.

Is it still a free license?
Never asked to #RMS or #FSF, but I guess that such formal constraint makes it "not free" to their eyes.

What they miss, imho, is that freedom without communion is always going to be exploited by the strongers (under #capitalism, the rich) to oppress the weakest (everybody else, the workers, the customers, the environment...) as #LLM are showing these days.

In fact the latest version of the #HackingLicense was written in response ti #GitHub #Copilot (aka #CopyALot), after it distributed #GPLv3 code from #Quake with a wrong attribution and a permissive license.

The Hacking License is a dependency inversion: if you use data or code covered by it, anything that come out can be used under such license.
The License-review Archives

On my way to the Magic of Open Source with @apell and @OSBA at the European Digital Summit in Berlin today- let’s talk about IT Excellence with Open Source! #linux #opensource #osaid #digitalsovereignty
@[email protected]

The way #LLM are voiding classic #copyleft has been experimentally demonstrated years ago, when #GitHub #Copilot was caught distributing well known #AGPLv3 code with a permissive license and wrong author attribution.

Also, technically that incident demonstrated how Copilot's model itself is a derivative work of copylefted works: if a lossy compression of copyrighted material is still subject to authors' #copyright, encoding such compression as arrays of floats that can be executed by virtual processors with a dedicated architecture (so called "inference" engines) does't change its nature of derivative work. Similarly violating copyrights of millions of authors at once doesn't free you of such rights.
That's basically why #OpenAI and friends are so scared by current lack of sustainable business models for their #LLM: they need money to keep Judges away.
But anyway I still have to find a single person that debate with technical competence and in good faith the derivative nature of LLMs from the text corpora compressed in their models.

As for the Hacking License not being a Free Software license, it's debatable after a careful read since the only thing you cannot do with the software is to prevent others from enjoining the same freedom it grants you.

Yet I've never claimed it is Free Software because, sadly, I'm forced to move beyond Free Software by its own limit.

OTOH I'm proud that it's not an #OpenSource license as I'll never submit it to #OSI corrupted #gatekeepers

As for #FLOSS, it's a term designed to confuse free software values with corporate propaganda while marginalizing hackers: its a leaking abstraction designed to fool developers and exploit their naive groupthinking. Having been fooled myself. Never again. 😉

____

¹ The way OSI tried to #openwash the #OSAID, with an over complicated process that doubled #Meta's lobbyists' votes to exclude training data from the requirements confirmed my opinion about them.

@[email protected]
GitHub CopyALot distributes GPL code with wrong attribution and a permissive license

PeerTube