Re: Coding assistants (LLM) policy

On Mon, 17 Aug 2026, Vivien Kraus <[email protected]> wrote:
> Hello,
>
> Le lundi 17 août 2026 à 10:22 -0400, Olivier Dion a écrit :
>>  If you do use LLM, we might never know.  In other words, we rely on
>>   the honor system.  If you do however find yourself using coding
>>   assistant, you must:
>> 
>>     - Review all the code yourself (no vibe code).
>> 
>>     - Ensure that the code is compliant with the licenses of the
>>       project.
>> 
>>     - You must add (or the committer will on your behalf if they
>> accept
>>       your change) a Signed-off-by tag to your commit to certify the
>>       Developer Certificate of Origin (DCO).  See the DCO document
>> for
>>       the terms you agree to.
>> 
>>     - The persons signing the commit (contributor and/or committer)
>>       take full responsibility for the contribution.
> With Anthropic about to implement watermarking to its text output, the
> rest of the industry may follow. We do not know much about it, but it
> is possible that in the near future, code generated by most LLMs will
> encode more or less secret metadata. What do you think about this?

Some have mentioned (here and on Codeberg) that LLM outputs and
copyright have no jurisprudence, yet.  Therefore, following the
precautionary principle, we need to assume the worst scenario.  I think
that could be in the form of the companies owning these models to hold
the copyright of the outputs.  And this is where the watermarking is
even more dangerous, because now, these companies will train their
models to inject payloads that are hidden and could be used in court.
Maybe I am totally off the track here and perhaps paranoid and not seing
the whole picture.

> Should the contributor disclose it, try and remove it, make it
> obvious, or something else?

I put in the policy the words "highly discouraged".  So I assume you
mean here that it would be okay perhaps to use LLM if disclosed and the
committers would use its jugment to accept or not the contribution.
After reading some the answers of others and thinking about it, I want
to expand on that a little.

I have avoided putting a straight up ban on the usage because the
question is more complex than good and bad.  Furtheremore, I wanted to
stay nuance to avoid any kind of stigmatization.  The topic of LLM is
indeed highly controversial in our field -- I would argue that this is
not the case for artists or other domains, peoples tend to realize how
LLM is not something they want/need in their life, at least in my circle
--, and even taboo for some.  Thus, I really want to avoid making a
them/us duality.

Thanks,
Olivier
-- 
Olivier Dion