How to deal with fully LLM-generated merge requests

[Vlad Zahorodnii]

Hello,

On 9/10/26 10:36 AM, Lasath Fernando wrote:
> I personally am a big fan of Jellyfin's AI/LLM use policy:
> https://jellyfin.org/docs/general/contributing/llm-policies/
>
> It's very pragmatic and somewhat enforceable. We should adapt something similar.

Yeah, it looks pretty reasonable.

Vlad

> Cheers,
> Lasath
>
> On Thu, Sep 10, 2026, at 02:27, Vlad Zahorodnii wrote:
>> Hi,
>>
>> We started seeing an increase in number of merge requests that have
>> obvious signs of being generated fully by LLMs. For example, walls of
>> text in merge description, or replying with a wall of text to a review
>> comment in less than a minute, etc.
>>
>> It is easy generating such merge requests, but going through them takes
>> time for us, humans. It's possible to encounter a denial of service type
>> of issue if the number of such MRs continues increasing.
>>
>> Personally, if I talk to a contributor, I would like to know that they
>> know what they do, that whoever uploaded the MR is not just a proxy
>> between me and their LLM. Lastly, I think it's rude to copy paste LLM
>> responses to comments made by humans.
>>
>> I suggest that we update our guidelines to say that it is okay to close
>> merge requests that have clear signs of being fully generated by LLMs
>> right away.
>>
>> It's also interesting what systemd and NetworkManager did recently
>> (https://www.phoronix.com/news/NetworkManager-AI-Canary and
>> https://www.phoronix.com/news/systemd-262-rc2). I don't know how well
>> that works but if it's proven to be effective, maybe we can adopt that
>> idea too?
>>
>> Regards,
>> Vlad