Re: GCC AI Policy Announcement

Richard Biener via Gcc <[email protected]> Mon, 3 Aug 2026 09:20:16 +0200
Newsgroups gmane.comp.gcc.devel,gmane.comp.gcc.fortran
Message-ID <CAFiYyc0C2t_FWcv9Wcz-dr3KUakMsKgyhDOSggVGaZpBWMspnQ@mail.gmail.com>
On Sun, Aug 2, 2026 at 2:35=E2=80=AFPM Richard Kenner <[email protected]> =
wrote:
>
> > That is one of the clearer points right now - LLM-generated code cannot
> > be copyrighted.
> >
> > If somebody significantly modifies LLM-generated code, then that person
> > gains copyright (which would also be fine).
>
> Yes to both.
>
> I want to point out that there has been and still is a long discussion
> on gnu-prog-discuss about this topic that's spanned many hundreds of
> messages.  I suggest that we not repeat that discussion here.
>
> The jist of that discussion is that there's concern that because
> LLM-generated code isn't copyrightable, there might become an issue
> where the GPL no longer applies because there's not enough
> copyrightable content.  The other side of that is the belief that there's
> a significant amount of human interaction in editing the resulting code,
> choosing the prompts, and deciding what to put where and that amount is
> enough to preserve copyrightability.
>
> The key problem, though, is that there's no legal precedents here and
> won't be for many years.  All court decisions even remotely related to
> this issue were decades before LLMs existed.  The US Copyright office
> has issued guidelines for AI in art, but not software, and with the
> overturning of Chevron, it's not clear that their guidance would have
> much legal authority anyway.
>
> So nobody, not even an attorney specializing in this field, can state
> with certainty which way this will go.  All anybody can do is
> speculate, based both on legal and political/business theories.  And
> that's an uncomfortable place to be.

I would like to get clarifications on the labelling requirement with
Assisted-By.
The policy talks about "including output" and "LLM generated content" but
avoids talking about derivative work of LLM generated content.  When I
use LLM to prototype an idea and then go a) implement the result myself,
b) edit the result [significantly], c) use it to create an improved
implementation
this is clearly "assisted by an LLM", does it require labelling with
Assisted-By?
I think the same legal issues are present?

I'd also guess that a corporate lawyer would advise against any labelling
with "Assisted-By" given the uncertainty in the legal realm.

Richard.