SCOTUS Refuses to Hear LLM-Backed AI Case Regarding Copyright - Conservation Blog

SCOTUS Refuse to hear LLM-Backed AI copyright case

by Bradley M. Kuhn
on March 4, 2026

No serious implications for FOSS of SCOTUS‘ Denial

Previously this week^{,

the US Supreme Court (SCOTUS) denied certiorari (certificate) in Thaler vs. Perlmutter. Thaler argued that an image—generated by a large language model (LLM)-assisted artificial intelligence (AI) — earned copyright registration. Since the US Office refused to grant the registration, Thaler appealed to the US District Court for the District of Columbia (DC circuit). This Court upheld the Office’s decision. SCOTUS’ denial of “sert” means that they will not hear the case. Strictly speaking, this denial does not confirmthe DC Circuit Court’s ruling, but it does mean that the DC Circuit decision stands.}

Many in the free and open source software (FOSS) community has expressed concern about the impact on copyleft – and even FOSS in general.
TL; DR: Don’t panic! —this case is extremely limited in scope.

First, a notice: this case is about copyright of an artistic image, not software. law – and the legal precedents surrounding it – vary widely for different types of creative works. Analysis of the copyright of works of software differs in notable ways. So don’t assume that analysis for images applies broadly to software.

Second, while the decision is “published”.²there is also many other cases relate to
LLMs and AI currently pending in the US courts. Courts and laws always lag behind technological advances. Indeed, that’s exactly why copyleft was invented: as a mechanism to achieve
existing laws and precedents that we could not achieve in the legislature. Forty-one years after copyleft’s invention, we still don’t have a federal law mandating software to restore!

Thirdly, the Court held that a registration was not valid (at this time) if the work’s sole author was a computer program. Thaler (who was both (one of) the author(s) of that computer program and its user) repeatedly waived any claim to consider Thaler’s own copyright in the LLM-supported AI incentive process. Nor did Thaler argue that any copyright interest in the LLM-supported AI system itself was subject to the registration. Thus, this decision does not evaluate any creative expression by (a) the author(s) of the incentives themselves, (b) copyright held in the LLM, its weights, generation, composition or its user interface, and (c) copyright held in underlying works in the LLM training data.

Thaler’s original registration was the main cause of this significant narrowing because the registration argued that the AI system itself was the author of the image. This case only consider a copyright registration where the sole “author” is identified as a specific computer program. Thaler determined that the work was generated only by incentives and no human changed the work afterwards. As such, even though the other districts begin to frequently cite this case, and even though many districts decide it applies to software without further consideration of the difference in the types of works, such a precedent does not spell disaster for
FOSS.

Admittedly, some LLM-backed generative AI agents may be simply asked to create a work of software from scratch which has some passing utility. The most common workflow in using these agents (at least in FOSS development) is as follows:

Start with an existing large FOSS code base.

Ask the LLM-backed AI agent to generate various changes and improvements to that code base.

Apply creatively, human attempt to modify and refactor the output to produce a patch suitable for upstream.

This case – in addition to the non-consideration of software at all – don’t consider that third step.

Furthermore, it dictates ³— appearing in the DC Circuit decision — supports a conclusion that human action steps on that third
would constitutes creative expression – affixed in tangible medium – suitable for copyright registration. Indeed,
their pronunciation declare:

First of allthe human authorship requirement does not prohibit copyrighted works made by or with the aid of artificial intelligence. The rule only requires that the author of that work be a human — the person who created, operated, or used artificial intelligence — and not the machine itself.
—(Thaler vs. Perlmutter130 F.4ᵗʰ 1039, 1049 (DC Cir. 2025))

Given the current state of LLMs and AI, this rule – even if universally adopted – will not cause serious damage to FOSS.

The Court also indicated that these other issues are for a future time in another case. The DC Circuit readily admits that their ruling applies only to the state of AI systems at the time of writing^{. again quote from their statement:}

Of course, the (Thaler’s AI) machine does not represent the limits of human technical ingenuity when it comes to artificial intelligence. Humans may at some point produce creative non-humans… Science fiction is replete with examples of creative machines that far exceed the capabilities of current generative artificial intelligence. For example, Star Trek’s Data may be worse than ChatGPT at writing poetry, but Data’s intelligence is comparable to that of a human. See Star Trek: The Next Generation: “Schism” (Paramount telecast October 19, 1992) (“Felis catus is your taxonomic nomenclature, an endothermic quadruped, carnivorous by nature”). There will be plenty of time for Congress and the Office to address those issues as they arise.
—(Thaler130 F.4ᵗʰ at 1050)

I completely agree with that statement by the Court. I also thank Judge Millett very much for quoting one of my top ten favorites
ST: TNG episodes to support the court’s orders.

(As always,
CFS is not a law firm, IANALand
TINLA. )

I – and my colleagues at SFC – recognize this SCOTUS recently made other decisions on a variety of important social justice cases. Since SCOTUS’ decision to deny cert in this particular case is so closely related to my work, I write about it. However, all of us at SFC recognize that our community is reeling under other recent decisions.

In this context, “published” is a term of art lawyers use to describe a case that the publishing court (in this case, the DC Circuit) deemed important enough to “officially and formally” share with other courts. While (in a precedent-based legal system) any Court dog citing an unpublished case from another Court, published cases are much more likely to be cited than unpublished cases.

^{“Dicta” is explanatory language found in a court’s decision that is not necessary to the court’s conclusion. . Dicta is not precedential, but it can be persuasive.}

^{Note that the DC Circuit issued their decision in March 2025. This is not unusual for

SCOTUS to delay for a year (or more) before issuing a decision to grant or refuse a certificate.}

(permalink)

Markers:

conservation area,
GPL,
law