Banning LLM-generated or assisted contributions to the GNU C Library

[Mark Wielaard]

Hi Andreas,

On Sat, Aug 23, 2025 at 01:32:18PM +0200, Andreas K. Huettel wrote:
> some time ago Gentoo Linux debated and decided an "AI Policy" [1]. Its verbatim text
> is 
> 
> > It is expressly forbidden to contribute to Gentoo any content that has been created with the 
> > assistance of Natural Language Processing artificial intelligence tools. This motion can be 
> > revisited, should a case been made over such a tool that does not pose copyright, ethical and 
> > quality concerns.
> 
> I would like to propose that the GNU C Library adopts a similar policy.
> 
> While I see the point of all rationales given in the Gentoo decision [1], my personal
> main drive is quality control and responsibility.
> 
> We are not talking about programming a mobile phone tower defense game here.
> We are however talking about code that is used in millions of devices worldwide in
> mission-critical and security-critical applications.
> 
> And for programming that, AI technology is not ready yet.
> 
> The GNU C Library would not be the second or third project to adopt such a policy [2,3]
> (or should adopt one [4]).
> 
> This is probably a topic that should be discussed at the Cauldron in Porto.

I agree this would be a good topic to discuss at the glibc BoF at the
Cauldron, but lets make sure to bring the discussion back to the
mailinglist.

I also agree that it is good to be proactive here and get some policy
in place. It is not just the legal/copyright issue. There are many
questional, moral, social issues around some of these "tools". We
might not be seeing some of the bad issues inside glibc yet, but
looking at other projects they might soon be here (even ignoring the
global issues).

There might be "good" uses of these AI/LLM generators. And it might
indeed be hard to define good-vs-bad.
You can imagine that if we can get to something like
https://sfconservancy.org/activities/aspirational-statement-on-llm-generative-ai-for-programming.html
it could be beneficial. But we are far from that at the moment.

It might be possible to describe some of the most worrying attributes
of bad tools like chatgpt, claude, grok, gemini, etc. If as a
contributor you don't have any insight into what went into the model,
cannot be sure about comes out, and cannot trust the "reasoning" then
you shouldn't submit any of the output of such a tool.

Maybe we can add something to code of conduct to catch other "abuses"
of the process, e.g. to disallow fuzzer generated "bugs" where the
fuzzer results cannot simply be reproduced or it needs a totally
unrealistic environment to trigger the issue (glibc build in a way to
no normal distro would). Basically make it unacceptable to
(intentionally) waste reviewers time through your contributions.

Thanks,

Mark

> [1] https://wiki.gentoo.org/wiki/Project:Council/AI_policy
> [2] https://www.netbsd.org/developers/commit-guidelines.html
> [3] https://www.qemu.org/docs/master/devel/code-provenance.html#use-of-ai-content-generators
> [4] https://xcancel.com/spendergrsec/status/1958264076162998771#m
> 
> -- 
> Andreas K. Hüttel
> dilfridge@gentoo.org
> Gentoo Linux developer 
> (council, comrel, toolchain, base-system, perl, libreoffice)
> https://wiki.gentoo.org/wiki/User:Dilfridge
> 
>