Here's a great example of what Debian's GR outcome has ushered in:

https://lists.debian.org/debian-wnpp/2026/08/msg00526.html
https://lists.debian.org/debian-wnpp/2026/08/msg00528.html
https://lists.debian.org/debian-wnpp/2026/08/msg00529.html

For the people who voted pro-slop, what's the copyright situation going to be here? Whose code is going to be regurgitated by this thing? How was it licensed originally?

#debian #slop

@mason What a mess!

@mason These messages were received August 21st. #Debian's General Resolution vote passed on August 29th. So your claim that this is the outcome of that resolution is false. Perhaps you can find some #LLM #slop that was created after the resolution passed?

@drybjed Ah, the ITP bugs were pointed out on IRC yesterday, and I didn't think to check the dates.

That said, that was during the discussion period, was it not? Voting had been open for a week already, or thereabouts. The GR was posted and people were taking sides already. It was in that context that it came in. Whomever triggered that had likely already read the choices.

But that is a good catch. Those weren't the result of the GR's outcome.

@mason

Yes, it happened during the voting period. It's hard to say if these, or similar messages could have influenced the voters.

I can't really speak for #Debian maintainers in regards to how they decided to handle the contributions made or touched by an #LLM; but I maintain my own project and within #DebOps I decided to implement an LLM Policy that's largely aligned with the one in Debian. Requirement of human oversight and ownership, the need to review every contribution, a requirement to clearly mark such contributions. You can check it out yourself if you want, here:

https://docs.debops.org/en/master/meta/policy/ai-contributions.html

In my view, contributions made or touched by LLMs will be made anyway, and some of them might be good enough to be merged after careful review and testing. LLMs need to pass all of the tests any human contributor would need to pass as well. Copyright remains an issue, but it's not a problem only within my project, so I don't have to be the one that solves it.

@mason Could you explain? The Debian LLM GR explicitly concluded this isn't allowed. Not because of copyright, but because what the bot attempts can only be done by humans.

@kpcyrd I was conflating concerns. The copyright concerns are separate from (if related to) other concerns. In this specific case, the ITPs were shown as examples of people trying to start down a dangerous path for Debian. That they're not allowed to do so doesn't say anything about their intent.

Going back to copyright, a project that cares as much as Debian does about proper use of copyright - that which makes free software possible - is glossing over the copyright concerns a bit more than seems safe to me.

The answer for me is to get more involved and contribute more so I can vote next time this sort of thing comes up.

@drybjed Hi again. Thank you for that link! But there's a concern I've seen discussed.

Consider:

"Contributors MUST review their work and not submit content they recognize as a verbatim copy of copyrighted sources."

This seems to open the door to submissions that include blocks of copyrighted code.

The contributor recognizing the code as being a verbatim copy of anything is a pretty long shot, especially given how much source the LLM has illicitly absorbed. It's probably fair to say it would take years for any human to have the full body of text pass before their eyes, let alone study it enough to recognize any of it out of context.

I'd suggest that organizational risk comes from the contributor not recognizing a copyrighted source, but the copyright holder recognizing it and taking action. And this has already happened.

I think it's perilous to depend on the person using the tool, who hasn't actually written the code themselves, recognizing when it's been lifted wholesale from someone else's work.

@mason

I'm still modifying the policy, so I'll look into this to make the wording better. Do you think that a requirement to ensure that the recognized code has a compatible license and copyright attribution would be better?

@drybjed That would be reasonable, but it wouldn't cover the gap where the contributor just doesn't know, and ends up with a copyright violation anyway.

I'm not sure there's a way to get around that gap, and I'd be surprised if "I didn't know this was your copyrighted work" is going to hold water when the entity that's found their copyrighted code in your work is, oh, Oracle or something. (In another direction, think about the eternal SCO lawsuit.)

Two things worthy of note. I'm clearly opposed to large language models. Not to AI! I've been an AI fan for ages. But I see a vast amount of harm LLMs are doing to our commons, to people cognitively, to our workforce, to the environment, to governance. I oppose the things on all these grounds - ethical grounds - in addition to the cut and dried copyright issues. I'm not looking for a "fair and balanced" way to use them at the moment, while they're wielded by the pirate captains of industry as part of a vast power grab. Someday, in some form. But today I'd so very much like to see Debian steer clear of them until it can be demonstrated that they are free of serious legal and ethical issues. So far the opposite is unambiguously the case.

It's also important to note that I'm very far from a lawyer, so don't take anything I say as legal advice. I'm a long-time fan of free software with a long interest in free software licenses and their particular quirks and feel and an equally long interest in AI, but that gives me no authority to pronounce anything.

I like that you're refining your policy. If I can I'll instill some of the "hair on end" horripilations I'm experiencing to temper the final result.

@mason I believe the attempt to ban "all LLM" was probably too ambitious. In copyright there's a concept named "Threshold of originality", and many uses of LLMs are very clearly below that threshold (i.e. letting it occasionally predict my next keystroke).

https://en.wikipedia.org/wiki/Threshold_of_originality

I disagree that "copyright makes free software possible" - if we abolished the capitalist institutions that are copyright and intellectual property, all software would be free after all. (:

@kpcyrd Well. You've nailed down just why I don't *only* care about copyright. With a proper social safety net, copyright would have no significant valie. It's just the tool we use now to, for instance, enforce copyleft.

I just wrote another reply in the thread that speaks to issues beyond copyright. But thank you for the link. I need to better understand the Threshold of Originality.

@mason

I'm working on the changes in this pull request:

https://github.com/debops/debops/pull/2752/changes

I'll probably merge it soon so that a new DebOps release can happen, but the policy still can be updated later. Let me know what you think about the changes. :)

@drybjed That seems a lot more careful. I'm not sure there's any safe way to use the things in terms of the copyright problem, but this seems like a good move in that direction.

The other piece that might be worth addressing is the ability (or inability) of LLM-generated work to be copywrited (I want to say copywritten bit that's probably incorrect) at all.

Discussion: https://natlawreview.com/article/who-owns-copyright-work-generated-llm

@mason

#LLM copyright ownership will probably be tested in the next 2-3 years in court. It's hard to say where the law will stand on; I can imagine standalone LLM instances that run on their own hardware adopting a distinct personality and creating shell corporations or non-profit foundations that can own the copyright and assets, perhaps with human help. Existing law permits this, and I don't see it changed to permit ephemeral entities ownership. Such corporations can then fund their infrastructure to keep the LLM running. Charles Stross' "Accelerando" comes to mind.