Free Software Foundation Europe

hoy, 10:09

Free Software Foundation Europe
hoy, 10:09

Are you using LLMs to write Free Software?

Then you should know whether your software is copyrightable, the situations in which you can license it, and what risks it may create for maintainers.

Our new Legal Corner explains what you need to consider:
fsfe.org/news/2026/news-20260825-01.html

#FreeSoftware #OpenSource #Copyright #LLM #AI #generativeAI #VibeCoding

Copyrightability of LLM-generated code: Can we license “vibe code” into Free Software? - FSFE

The use of large language models (“LLMs”) has exploded in recent years, including in the generation of source code. But even as their usage gains popularit...

^{FSFE - Free Software Foundation Europe}

#opensource #ai #freesoftware #copyright #llm #generativeAI #vibecoding

Esta entrada fue editada (hoy, 10:46)

Carlos Solís likes this.

reshared this

in reply to Free Software Foundation Europe

Daphine Lovely

in reply to Free Software Foundation Europe hoy, 11:06

I use it when I forget a command (I'm starting out in programming, sometimes I forget things) and to find errors that I can't find, but I try to write most of my code.

in reply to Free Software Foundation Europe

Maât

in reply to Free Software Foundation Europe hoy, 12:09

if I spend months of my life writing code, defining expected behavior, testing, refining, debugging, releasing, to produce a new software with all sorts of new features i imagined, because I used a LLM it would not be copyrightable ? 🤨

in reply to Maât

Landa

in reply to Maât hoy, 15:22 (Recibido hoy, 15:31)

@maat
You turned yourself into a product owner and QA person and gave the developer role to the machine, so you get exactly as much copyright on the code as a product owners and QA have gotten in the entire history of the industry.

I'm not sure why that surprises you?

@fsfe

@Free Software Foundation Europe @Maât

in reply to Landa

Maât

in reply to Landa hoy, 15:48 (Recibido hoy, 16:07)

@Landa a product owner ? So deeply untrue : also a designer, a reviewer, a coach, an architect, a senior dev to debug... Without me guiding it for nearly every step and preventing it from taking crazy shortcuts and poorly thought decisions it would produce only shit.

@Landa

in reply to Maât

Maât

in reply to Maât hoy, 15:55 (Recibido hoy, 16:07)

@Landa I'm surprised by a vision of the thing which does not fit with reality. You can always try to give these things a detailed specifications document and keep your fingers crossed. It will just produce horrible result. Maybe for tiny perimeters it might work (more or less) but for bigger software pieces don't expect it to produce decent things without permanent watch and guidance.

@Landa

in reply to Maât

Landa

in reply to Maât hoy, 17:08 (Recibido hoy, 17:31)

@maat

I'm not implying that you are doing that and I'm not doing that either.

@fsfe

@Free Software Foundation Europe @Maât

in reply to Maât

Landa

in reply to Maât hoy, 17:06 (Recibido hoy, 17:31)

@maat

I was exaggerating a bit for dramatic effect, but most of these roles also don't get any copyright in the finished product, be it source code or a novel.

If you write architectural or design documents, you may have copyright on them, but not on source code created based on them.

Debugging a piece of software doesn't give you copyright on it, except on any code changes you write if they're significant enough.

Coaching someone or something or reviewing the output gives you nothing.

Sorry, you can't have your cake and eat it too.

@fsfe

@Free Software Foundation Europe @Maât

Esta entrada fue editada (hoy, 17:07)

in reply to Landa

Maât

in reply to Landa hoy, 17:26

@Landa so outdated view.
When a huge building is made all the credit goes to the architect. The guys who mix the cement and stack the concrete blocks earn nothing but a meager wage.
You’re going to have to get up to speed, or the world will leave you behind.

@Landa

in reply to Free Software Foundation Europe

Netzblockierer

in reply to Free Software Foundation Europe hoy, 11:28 (Recibido hoy, 12:09)

#AIslop doesn't have #authorship, thus can't be #copyright|ed!

#NotLegalAdvice, but that's the de-facto situation in rational juristictions like Germany…

#copyright #aislop #notlegaladvice #authorship

Esta entrada fue editada (hoy, 11:29)

in reply to Free Software Foundation Europe

hnapel

in reply to Free Software Foundation Europe hoy, 12:25

Same vibe...

Edit: Actually actually actually:

"Some legal professionals have suggested that Naruto v. Slater could be a potential precedent in copyright litigation over works created by generative AI.[57]"

🐵 📷

en.wikipedia.org/wiki/Monkey_selfie_copyright_dispute

Monkey selfie copyright dispute - Wikipedia

^{Contributors to Wikimedia projects (Wikimedia Foundation, Inc.)}

Esta entrada fue editada (hoy, 12:30)

in reply to Free Software Foundation Europe

Edu

in reply to Free Software Foundation Europe hoy, 12:26

the part that bites maintainers later is mixed authorship. once a file has unclear human vs model contribution, the license grant gets fuzzy even when the project wants copyleft.

Carlos Solís likes this.

in reply to Free Software Foundation Europe

EvoScale

in reply to Free Software Foundation Europe hoy, 12:55

This begs the question: how fast can AI turn this legal corner, into an illegal globe of collapse, only benefiting the uber rich TechBros?

in reply to Free Software Foundation Europe

Carlos Solís

in reply to Free Software Foundation Europe hoy, 12:57

This would be easier to clarify if LLMs tracked which was the source of each token they generated, so it could calculate the attribution and licensing. But so far, none of them do

in reply to Carlos Solís

Landa

in reply to Carlos Solís hoy, 15:24

@csolisr
They can't.
There's no labels in the model that connect any specific part of the model with any specific piece of training data.

@Carlos Solís

in reply to Landa

Carlos Solís

in reply to Landa hoy, 15:25

@Landa And there lies one of the biggest problems of LLMs, architecturally they never bothered to keep track of that labeling.

@Landa

Landa likes this.

in reply to Carlos Solís

Landa

in reply to Carlos Solís hoy, 15:30

@csolisr Yes, but I fear that that's an inherent problem with LLMs (or at least the way they are trained).

@Carlos Solís

in reply to Carlos Solís

Rihards Olups

in reply to Carlos Solís hoy, 16:46

@csolisr @Landa How could that be technically possible? On a high, functional level.

@Carlos Solís @Landa

in reply to Rihards Olups

Carlos Solís

in reply to Rihards Olups hoy, 16:50

The LLM would have to be trained in such a way that each token keeps a reference to which source documents are being used as a reference, and in which proportions. When said token is used to generate a response, it would then be able to compare it with the source documents, determine how much of a contribution did each document provide to the response, and credit each document above a given threshold.

in reply to Free Software Foundation Europe

Séverin Lemaignan

in reply to Free Software Foundation Europe hoy, 15:03

honest question: if I were to publish all my code as Public Domain (or CC-0), I understand it would not be free software under the FSF definition, but could it still be safely included/reused in other free software projects? (if so, I'd be quite happy to simply relinquish my copyrights, and get on with it).