Re: [MAINTAINERS SUMMIT] Other LLM-related topics - tags, newcomers, etc

On Fri, 2026-07-17 at 21:13 -0400, Theodore Tso wrote:
> On Fri, Jul 17, 2026 at 02:42:09PM -0500, James Bottomley wrote:
> > They're not really enforced at all.=C2=A0 It's the legal equivalent of =
a
> > representation, it just means I read the DCO and I represent I
> > conform to what it says.=C2=A0 Externally we have no verification
> > mechanism that the person signing off actually did this ...
> > although if they're regular contributors the trust factor is
> > higher, and if they get caught faking signoffs then the trust
> > factor goes down.
>=20
> This is something that perhaps we should ask legal council.=C2=A0 The DCO
> states (among other things):
>=20
> (a) The contribution was created in whole or in part by me and I
> =C2=A0=C2=A0=C2=A0 have the right to submit it under the open source lice=
nse
> =C2=A0=C2=A0=C2=A0 indicated in the file; or
>=20
> (b) The contribution is based upon previous work that, to the best
> =C2=A0=C2=A0=C2=A0 of my knowledge, is covered under an appropriate open =
source
> =C2=A0=C2=A0=C2=A0 license and I have the right under that license to sub=
mit that
> =C2=A0=C2=A0=C2=A0 work with modifications, whether created in whole or i=
n part
> =C2=A0=C2=A0=C2=A0 by me, under the same open source license (unless I am
> =C2=A0=C2=A0=C2=A0 permitted to submit under a different license), as ind=
icated
> =C2=A0=C2=A0=C2=A0 in the file; or ...

Yes, I know ... I spent a decade or so selling the DCO to the industry
as TAB chair.

> Given, however, that we don't know whether it might mean that
> vis-a-vis code which is generated by AI,

Well, we do currently: as the copyright office asserts AI can't hold
copyright in generated code then anyone can submit it satisfactorily
under the DCO because they have the right to do so in the same way
they'd have the right to do so for any public domain code.

However, just in case that advice changes it's not unreasonable to ask
the submitters to label the code as AI generated.

>  we might want to ask competent legal counsel questions such as:
>=20
> 1) Should we change the DCO to require disclosure if the code
> was generated by an LLM?

Realistically it's taken decades for industry and its legal counsels to
accept the DCO as standard practice, the foundation of which is the DCO
is simple and doesn't change; so suddenly updating it on an AI whim is
going to really upset that consensus and shouldn't be done.  That means
whatever we do for AI must be reasonable and not involve changing the
DCO.

> 2) If so, what detail is needed?=C2=A0 Does it matter, or potentially
> matter, what model(s) was used to generate the code?
>=20
> 3) Does it matter, or potentially matter, whether the LLM was used to
> generate code out of whole cloth, or to fix existing code?=C2=A0 For
> example, if the LLM inserts a spin_unlock() call, or modifies the
> prexisting code to add a scoped guard to avoid a memory leak, the
> chances that this might have copyright implication might be different
> compared if the LLM is asked to generate a new device driver for some
> graphics card.
>=20
> 4) Does the prompt or series of prompts that were used matter, or
> potentially matter?=C2=A0 If so, do we need to record them in the git
> commit description, much like we sometimes include the coccinelle
> script?

I think we can establish a project specific consensus on how we'd like
the labelling to be done.  That the contribution was created using AI
is the most important.  The model and prompts may or may not be
interesting, but we can likely find them as part of the historical
record.  Representations based on project norms are just as useful as
the ones based in the DCO (as we know from the plethora of tags we use
that have no basis in the DCO).

The only point I was making is that I think the AI generated tag has
future use as part of the code base and thus we shouldn't simply
eliminate it because reviewers don't find it useful.

Regards,

James