[RFC] glibc AI policy. Round 3. Adopt the GCC AI policy.

Carlos O'Donell carlos@redhat.com
Tue Nov 3 22:58:05 GMT 2026


On 10/28/26 5:30 AM, Jonathan Wakely wrote:
> On Tue, 27 Oct 2026 at 21:42, Andrew Pinski <pinskia@gmail.com> wrote:
>> I am saying the policy should have a recommendation on listing what
>> parts of the patch was asisted by the LLM (e.g. tests, review, finding
>> the issue in the first place, etc.).
> 
> I've been simply explaining it in the commit message, rather than
> trying to squeeze it into the Assisted-by: trailer, e.g.
> https://gcc.gnu.org/g:05ac72feea409f96c2b690a6f064171e274c8f91
> https://gcc.gnu.org/g:524847fec983eaedca147840ce46fcbb0cd38952

Agreed, this is the best way to describe it, particularly since how
the test was generated might matter to the context as Florian's example
below does e.g. based on tst-robust12.c.

Example:
https://sourceware.org/cgit/glibc/commit/?id=65b0a5da762671a54781044630e6c4f651312e88

With a human reviewer providing "Reviewed-by:" 
>> Note the assisted by tag I don't
>> think is needed/required to record for reviews fo the patch; though
>> that is a nice thing to have; though we have a "reviewed-by" tag now;
>> maybe `reviewed-by: LLM` is ok there rather than "assisted-by".
> 
> Why? I don't think "Reviewed-by: LLM" is useful. Who cares? The policy
> is clear that using an LLM for purposes like understanding the code or
> debugging is fine, only generating code or other content to be
> committed requires attribution.

Agreed, we need "Reviewed-by:" to mean human review.

-- 
Cheers,
Carlos.



More information about the Libc-alpha mailing list