glibc AI policy

[Florian Weimer]

* Carlos O'Donell:

> On 4/11/26 1:17 PM, Florian Weimer wrote:
>> * Sam James:
>> 
>>> Florian Weimer <fw@deneb.enyo.de> writes:
>>>
>>>> * Jeffrey Law:
>>>>
>>>>> On 4/2/2026 12:56 PM, Collin Funk wrote:
>>>>>> DJ Delorie <dj@redhat.com> writes:
>>>>>>
>>>>>>> Given that GNU is a USA-based project, and that the copyright status of
>>>>>>> AI in the USA is still... "nebulous at best"... it might take time to
>>>>>>> figure out whether or not we can accept that first AI patch that comes
>>>>>>> along.  Better to take that time now so we can respond more quickly
>>>>>>> then.
>>>>>> The copyright situation is the reason I am against tentatively accepting
>>>>>> patches while waiting for a more official policy from the FSF. Thanks
>>>>>> for mentioning that.'
>>>>
>>>>> Yea, that's where I'd land as well.  Until the copyright situation is
>>>>> better understood I would not accept any LLM generated code that is
>>>>> outside the de-minimis guidelines set by the FSF.
>>>>
>>>> Has the FSF communicated this (not accepting LLM-generated
>>>> contributions for now) to the glibc stewards?  Why would the
>>>> de-minimis guidelines apply, anyway?  I assume those are about not
>>>> diluting the copyright assignment, but that's hardly a concern for GNU
>>>> components accepting contributions under the DCO.
>>>
>>> Outside of communication to glibc stewards which I would not be party
>>> to, with regard to Emacs, there was
>>> https://lists.gnu.org/archive/html/emacs-devel/2026-03/msg00554.html.
>> Thanks for the reference.  This is what Richard Stallman wrote:
>> | The FSF is working on one.  Please wait for us to finish and
>> publish
>> | it before stirring up disputes about hypothetical options.
>> I assume the net effect that for Emacs, it's a case-by-case decision
>> if
>> a machine-generated contribution is accepted.  So it's pretty much like
>> what we do for glibc today, I think.
>
> I agree. Until we have such guidance I will review such submissions on a
> case-by-case basis. I have already approved one for Ajun Shankar
> e.g. 244c404ae85003f45aa491a50b6902655ee2df15 which has 6 boiler plate
> makefile lines generated by an LLM and were not legally significant.
> We did not add an "Assisted-by:" line for that change, but any such
> process that requires humans to remember "Assisted-by:" lines is going
> to have some missing compliance aspects as a function of cost of
> compliance e.g. Reviewed-by: etc.

For the record, I'm not happy with requiring Assisted-by:.  If those
companies want us to advertise their products, that's something they
should pay for.  It also raises the question whether we need to track if
users paid models or the free version, which may have different
ownership terms of generated materials.  Without such information, it
would be misleading to use the trailers for any project-specific purpose
(which is why I think it's mostly about advertising).

Thanks,
Florian