On 3/5/2026 3:53 PM, Greg Troxel wrote:
Jim Porter <[email protected]> writes:
The US Supreme Court has recently affirmed that the outputs of
generative AI aren't copyrightable; the "author" is held to be the
computer, but copyright is only applicable to creative works authored
by humans. The result is that the gen-AI outputs are public domain.
That's the US, and Free Software rests on a shared understanding of
copyright that applies broadly (across Berne Convention countries).
Did they really say that so precisely? I can certainly see that there
can be no original copyright in LLM output, lacking a human, but I don't
see how copyrighted input works are laundered of their copyrights.
Specifically, the US Supreme Court declined to hear an appeal in Thaler
v. Perlmutter[1], where the plaintiff argued that he should be allowed
to copyright an AI-generated image. As far as I know, the court didn't
address the question of whether it's legal to use copyrighted works as
inputs to train the models in the first place. (Or what happens if an
end-user prompts a model to regenerate an approximation of one of its
copyrighted inputs.)
I probably overstated matters when I said that gen-AI outputs are public
domain without any additional qualifications. I suppose that only
applies if the work isn't infringing copyright in the first place.
[1]
https://www.reuters.com/legal/government/us-supreme-court-declines-hear-dispute-over-copyrights-ai-generated-material-2026-03-02/
