On Oct 22, 2026 09:51, Andrei Ionuţ Ştefan <[email protected]>
wrote:
>
> Andreas,
>
> That doesn't actually answer the question. Having a COPYING file that says
> "GPL" just restates what license you're claiming — it doesn't answer either
> part of what was actually asked:
>
> 1. Can that claim legally hold at all, if a substantial part of the code is
> LLM-generated? LLM output's copyright status is unsettled — if there's no
> underlying copyright, there's nothing for a GPL declaration to attach to.
> Stating the license doesn't resolve that question, it just assumes the answer.
> 2. Has anything actually been run to check the code for verbatim reproduction
> of copyrighted training data? A license file in the repo says nothing about
> what's actually in the code underneath it.
>
> I'm not assuming there's a problem here — I'm asking because a
> mentors.debian.net reviewer or an ftpmaster during NEW review is going to ask
> exactly this, and "the COPYING file says so" won't be a satisfying answer
> there either. A real answer would be something like: an estimate of how much
> of the codebase is LLM-generated, and whether you've run a clone/license
> scanner (ScanCode, FOSSology) against it.
>
> Sincerely,
> Ştefan Andrei Ionuţ
I don't think you, or debian-legal will be able to settle the copyrightable
nature of LLM generated code.
Thomas
>>>
