On 2026-05-26 00:57, Andy Tai wrote:
The US Copyright Office declared that LLM output is non-copyrightable.
Is
LLM output public domain? Then incorporating LLM output into
copylefted
program seems safe, as such output is GPL'd.
Let us be very specific and within the context.
Your statement "The US Copyright Office declared that LLM output is
non-copyrightable." is wrong, it is in the wrong context and without the
context!
Telling like that generally to public reading it will make some of them
even believe it!
Thus I suggest you to read and understand the context:
Copyrightability of genAI outputs in the US: Key developments | DLA
Piper:
https://www.dlapiper.com/en-gb/insights/publications/2025/03/copyrightability-of-genai-outputs-in-the-us-key-developments
Artificial Intelligence and the Public Domain | Copyright Corner:
https://library.osu.edu/site/copyright/2026/02/06/artificial-intelligence-and-the-public-domain/
The result of the analysis is that your statement should be corrected to
following:
Incorrect statement:
"The US Copyright Office declared that LLM output is non-copyrightable."
Corrected statement:
"The U.S. Copyright Office has clarified that strictly AI-generated
works—those created solely by a machine without sufficient human
intervention—are not eligible for copyright protection and are therefore
in the public domain. However, if a human adds creative elements to the
AI output, such as arrangements or modifications, then they may claim
copyright in their contributions. The mere provision of prompts does not
provide the requisite human authorship for copyrightability."
Because your first statement sounded as truth believed by you, now it is
questionable how much the rest of the statement really apply!
But there seems to be fear that if much of a program is output from LLM
(say via gradual patch incorporation or third-party contribution), then
the
copyright of the whole program—GPL or some other license—may no longer
apply,
I'm curious if such fear is justified.
But let me say short: That specific fear is not justified under current
U.S. law.
A program that contains LLM-generated output does not lose its overall
copyright, even if most of the code comes from an LLM. The GPL (or any
other license) still applies to the human-authored parts, and the
program as a whole remains protected.
Fear is not warranted because copyright is about human authorship, no
matter which tools you use. It is not about the medium or proportion.
Pure LLM output (no human creative input) is uncopyrightable → public
domain.
Human modifications, selection, arrangement, or integration of that
output are copyrightable.
Even if 99% of the code came from an LLM, the 1% of human decisions (how
to combine patches, which outputs to keep, how to structure the program,
fixing bugs, adding logic) are protectable.
Copyright does not vanish just because the uncopyrightable portion is
large.
The whole program is not “contaminated” by public domain material!
Courts have long held that a work containing uncopyrightable elements
(facts, ideas, public domain text) can still be copyrighted as a
compilation or derivative work, provided there is sufficient human
authorship in the selection, coordination, or arrangement.
Example:
A GPL-licensed program that includes a public domain sorting function is
still a copyrighted GPL program.
The same applies to LLM-generated code: it’s just public domain material
you incorporated.
The GPL remains enforceable on the human-contributed parts
The GPL requires you to license the whole work under the GPL if you
distribute it. That is still possible because:
You (the human) own the copyright to your original contributions.
The public domain LLM portions have no owner, so they don’t block the
GPL.
The FSF has explicitly stated that combining public domain code with
GPL’d code is allowed, and the result is still a GPL’d work.
--
Jean Louis
_______________________________________________
libreplanet-discuss mailing list
[email protected]
https://lists.libreplanet.org/mailman/listinfo/libreplanet-discuss
