[gdal-dev] RFC111 AI/LLM tool policy: proposal for a significant revision

[Greg Troxel]

Even Rouault <even.rouault at spatialys.com> writes:

> Le 06/05/2026 à 19:49, Greg Troxel a écrit :
>> Even:
>>
>>    I would somehow keep the idea that it is not settled law if LLM output
>>    is a derived work of the inputs, in the discussion of legal issues.  I
>>    keep hearing that people are able to get LLMs to output training data
>>    verbatim, proving it's in the model,
> Please make a text suggestion in the PR if you want that point to be
> captured. Independently of whether the LLM would output verbatim, even
> if it is "creative", it looks like the legal opinion would be that it
> isn't copyrightable, because copyright is linked to human
> production. You could perhaps copyright the prompt (not sure, but why
> not? if it is creative enough), but not what the LLM has spit out from
> it.

Will do.

I agree that the output itself is not eligible for copyright, because it
was not created by a human.  There is a separate question of whether it
is a derived work of other copyrighted works, and thus if one needs
permission under copyright law from those authors.

consider a prompt: take the first 4 paragraphs of [famous book in
copyright] and add two more that tlak about foo


>>    Other than that, your revisions sound good to me.  If it turns out
>>    other projects figure out a good way to interact and the legal issues
>>    are resolved, this can be changed later.
>
> My main issue is not about the legal aspect, more about the weird
> interaction process of having a machine generating code vs a human
> reviewing it. I truly believe that if you accept LLM generated code,
> then in practice, you must also give up on human reviewing.

100% agreed.

I was merely pointing out that saying no now is not forever, and it's
not dangerous to say no.