Second movement on origin: what the law truly seeks when it searches for an author
In a previous text, I argued that we possess a cultural obsession with origin: we wish to know the provenance of a work, whose hand guided it, and what process produced it, as if its value depended upon the purity of its source. That text examined the cultural facet of this obsession. I now intend to examine the legal facet, for therein lies a phenomenon that illuminates the entire problem from a different perspective: the law is also obsessed with origin, yet not for the reasons we typically assume.
The case that has made this clear has a name and a date. Stephen Thaler attempted to register a work, A Recent Entrance to Paradise, declaring his artificial intelligence system as the author. The copyright office denied it. The courts confirmed the denial: first a federal district court, then the Court of Appeals for the District of Columbia, which on 18 March 2025 established the criterion—the Copyright Act requires human authorship in the first instance. On 2 March 2026, the Supreme Court declined to review the case. Precision is required: the Supreme Court did not issue a ruling on the merits; by refusing to hear the case, it effectively left the lower court's decision standing. The finality is provided by the District of Columbia; the Supreme Court merely stepped aside. And the resulting legal conclusion is narrow and firm: a work whose only identified author is a machine is not registrable. Every protectable work must have been created, in the first instance, by a human being.
1. The easy reading, and why it falls short
The immediate interpretation of Thaler is humanistic and reassuring: the law defends human creativity against the machine, draws a line, and protects what is ours. Such a reading is not false, yet it is superficial, and it is prudent not to remain there, for it attributes to the law an intention it does not possess.
The law does not demand a human author out of devotion to creativity or romantic nostalgia for genius. It demands one for a reason far more prosaic and revealing: it requires a point of imputation. Someone to whom the right may be attributed, certainly, but also the limitations, the obligations, and the liability. Copyright is not a medal for creative merit: it is a system for the assignment of ownership, and a system of ownership requires an owner who can be held accountable. A machine cannot be sued, cannot assign rights, cannot assume liability, and cannot die so that the work may enter the public domain. It is not that the law considers AI unworthy of authorship. It is that the category of author, for the law, is primarily a category of imputation, and a machine is not imputable.
Viewed in this light, Thaler does not defend the human artist. It resolves a matter of legal plumbing: where to anchor the chain of rights and responsibilities. And in resolving it, it exposes the true problem.
2. The problem that Thaler opens
For Thaler closes the most extreme avenue—AI as the sole author of an autonomously generated work—and, in closing it, opens the question that truly matters: what degree and what type of human intervention render a work created with AI attributable to a person?
The United States Copyright Office has already begun to respond, and its answer is the decisive piece. In its report on copyright and artificial intelligence, it maintains that generative results may be protected only when a human author has determined sufficient expressive elements. And it specifies something crucial: the mere introduction of prompts is insufficient. Writing an instruction, however elaborate it may be, does not automatically render the writer the author of the result, because between the instruction and the image lies a margin that the machine fills in a manner the human does not entirely control.
Therein lies the shift that interests me, and it is the thesis of this text. Authorship, following generative AI, does not shift from the hand to the machine. It shifts from the hand to control. What the law seeks is no longer the physical trace of the creator—the stroke, the gesture, the hand that held the brush—because that trace has vanished. It seeks something else: sufficient decision-making, expressive control, the selection of one possibility among many, the intervention that transforms an automatic result into a form imputable to someone. Not who pressed the button, but who governed the process.
3. Europe, by a different path, to the same place
It is prudent to observe Europe, for it confirms the pattern via a different path. In the European Union, there is no specific rule that grants copyright to works autonomously generated by AI without human contribution. The European approach pivots on its own standard, developed by its court of justice: a protectable work is one that constitutes an “author’s own intellectual creation,” an expression of the free and creative decisions of a person.
The language is different, the legal architecture is different, but the vector is the same: human-centric. The United States formulates it as a requirement of human authorship; Europe, as a requirement of one’s own intellectual creation. Where they converge is at the minimum point—the machine is not an author—and, more interestingly, in the manner of managing the boundary: both systems ultimately seek, in generative work, not the hand but sufficient human decision-making. The European AI Act, for its part, adds obligations of transparency and governance, yet it does not convert generative results into protected works. It does not resolve authorship; it leaves it, like Thaler, in the hands of the question of control.
4. What this tells art
The origin, therefore, does not vanish with generative work. It becomes more technical. When an image no longer bears the evident trace of a hand, the law—and, behind it, a significant portion of our intuition regarding authorship—seeks another trace: that of the governance of the process. Who decided, who selected, who discarded, who intervened sufficiently for the result to be attributable to them. The obsession with origin is not cured; it merely changes its object. It shifts from the hand to the chain of decisions.
This has a consequence that extends beyond the law and touches upon what I have been considering regarding art. If authorship is measured by control rather than manual trace, then the artist who works with generative systems is no less an author for not having touched the image by hand: they are an author to the extent that they govern the process, configure the conditions, select from what the system offers, and intervene until they produce a form that is their own. The hand ceases to be the criterion. The decision is. And this, far from degrading the figure of the artist, describes it more accurately: the artist was never merely a skilled hand; they were always, above all, someone who decides, who chooses one possibility and discards a thousand, who governs a process until something becomes form.
There remains, however, an area that neither the law nor this reflection fully resolves, and it is prudent not to feign that it is closed: how much control is “sufficient” has no fixed answer. There is no stable methodology to quantify how much human decision-making is required for a generative result to constitute a work. That indeterminacy is real and will give rise to litigation for years. Yet it is an honest indeterminacy: it acknowledges that authorship was always a matter of degree, not a clear threshold, and that generative AI has not created this problem, it has merely made it impossible to ignore.
There is, in the end, something almost liberating in this displacement. For centuries we confused authorship with manual dexterity, and that confusion excluded much art that did not pass through it. That the criterion now shifts toward the governance of the process returns authorship to where it truly always resided: in the decision, in the gaze that chooses, in the will that gives form. The hand that signs may disappear. The decision that governs, does not. And as long as there is someone deciding what deserves to exist among everything a machine can produce, there will continue to be an author, even if there is no hand.
On open conversation
This text is the second movement of The Obsession with Origin and intersects with the reflection on provenance and infrastructure that I explore in this Notebook. If anyone wishes to intervene from the perspective of copyright law, artistic practice with generative systems, or the philosophy of authorship, the notebook remains open.
Sources
United States Court of Appeals for the District of Columbia Circuit (2025). Stephen Thaler v. Shira Perlmutter, No. 23-5233, decision of 18 March 2025. https://media.cadc.uscourts.gov/opinions/docs/2025/03/23-5233.pdf
Supreme Court of the United States (2026). Denial of certiorari in Thaler v. Perlmutter, docket No. 25-449, Order List of 2 March 2026. https://www.supremecourt.gov/docket/docketfiles/html/public/25-449.html
U.S. Copyright Office (2025). Copyright and Artificial Intelligence, Part 2: Copyrightability. 29 January 2025. https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf
European Parliamentary Research Service (2025). Karttunen, Sofia. Copyright of AI-generated works: Approaches in the EU and beyond. PE 782.585, December 2025. https://www.europarl.europa.eu/RegData/etudes/BRIE/2025/782585/EPRS_BRI%282025%29782585_EN.pdf
Fritz, Johannes (2025). “Understanding authorship in Artificial Intelligence-assisted works.” Journal of Intellectual Property Law & Practice, 20(5), 354-364. https://doi.org/10.1093/jiplp/jpae119
European Union (2024). Regulation (EU) 2024/1689 (AI Act). http://data.europa.eu/eli/reg/2024/1689/oj