A computer scientist named Stephen Thaler once filed a copyright application for an image and listed his AI system as the sole author, with himself credited only as the machine's owner. He wanted the software recognized, not himself. The Copyright Office said no. A federal district court agreed in 2023. The D.C. Circuit agreed again in 2025. In March 2026 the Supreme Court declined to even hear the argument. Thaler lost at every stage, and he was arguably making the simplest possible version of the case: he owned the system outright and asked for nothing but the paperwork to say so.
That case is the floor. The far more common situation, an author who used AI to help draft part of a book, textbook, or monograph and now wants to register the copyright, is harder, and the US Copyright Office has spent the past three years building out exactly where the line falls. The answer matters beyond paperwork: registration is what lets you sue over copying in US federal court, so it decides, concretely, which sentences of an AI-assisted work you can actually defend.
The floor: Thaler v. Perlmutter
The D.C. Circuit's holding, affirmed March 18, 2025, leaves little room to negotiate:
"The Creativity Machine cannot be the recognized author of a copyrighted work because the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being."
Thaler v. Perlmutter, D.C. Circuit, No. 23-5233
On March 2, 2026, the Supreme Court denied certiorari, leaving that rule in place as settled law: work with zero human authorship gets zero copyright protection, regardless of who owns or built the system that generated it (D.C. Circuit opinion, No. 23-5233; ReedSmith, on the cert denial).
The middle: Zarya of the Dawn
Zarya of the Dawn is a graphic novel with human-written text and AI-generated (Midjourney) illustrations. Rather than an all-or-nothing outcome, the Copyright Office split the work in its February 2023 decision: it protected the human-written text and the human "selection, coordination, and arrangement" of the whole work, while denying protection to the individual AI-generated images themselves (case analysis, University of Maryland Journal of Business & Technology Law). The precedent this sets matters more for written work than the Thaler case does: a single document can be partially protected, with protection tracking which specific elements a human actually authored.
The operative standard: Copyright and Artificial Intelligence, Part 2
The Copyright Office's report on copyrightability, published January 29, 2025 after review of more than 10,000 public comments, is the current operative standard:
"Copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements. Based on the functioning of current generally available technology, prompts do not alone provide sufficient control."
Copyright and Artificial Intelligence, Part 2: Copyrightability, US Copyright Office
Writing a prompt and accepting the first output is not enough. Selecting a favorite among several AI-generated outputs is not enough either: selection alone is not a creative act. What does qualify, per the same report: "Human authors are entitled to copyright in their works of authorship that are perceptible in AI-generated outputs, as well as the creative selection, coordination, or arrangement of material in the outputs, or creative modifications of the outputs" (Copyright and Artificial Intelligence, Part 2, US Copyright Office; official Copyright and AI hub).
Critically, the report also restates a disclosure obligation first set out in the Office's 2023 registration guidance: "if a work contains more than a de minimis amount of AI-generated material, the applicant should disclose that information and provide a brief statement describing the human author's contribution."
Four decisions, laid out chronologically
| Case or report | Date | What it established |
|---|---|---|
| Zarya of the Dawn decision | February 2023 | A single work can be split: human-written text and arrangement protected, AI-generated images not |
| Thaler v. Perlmutter, D.C. Circuit | March 2025 | Zero human authorship means zero copyright, regardless of who owns the generating system |
| Copyright and AI, Part 2 | January 29, 2025 | AI output is copyrightable only where a human determined sufficient expressive elements; prompting alone is not enough |
| Supreme Court denies cert in Thaler | March 2, 2026 | The human-only rule stands as settled law |
What "describe the human contribution" actually asks for
This disclosure obligation is a heavier evidentiary bar than a journal's Acknowledgments paragraph. An applicant is making a specific, sworn claim to a federal office about which words, structures, and edits are theirs, under the same penalties that attach to any other misrepresentation in a copyright application. If that claim is later shown to be inaccurate, whether through carelessness or convenience, the consequence is not a correction: it is a fraudulent-registration problem. And most authors, asked months after finishing a manuscript to specify exactly which passages they typed directly versus accepted from an AI suggestion, are working from memory of a process nobody recorded at the time.
An account that doesn't rely on memory
Write the manuscript in TeXposit and there's already a ready-made version of the account the Office is asking for: a per-word, per-contributor breakdown of which passages were typed directly and which came from an accepted AI suggestion, drawn from the document's real edit history rather than whatever the author remembers months later. The provenance report doesn't decide copyrightability. That's a judgment for the Office or a court to make about the finished expression, not about a report describing how it was produced. What it changes is the shape of the claim an applicant can make: not "I mostly wrote it, AI just helped with some sections," defensible only until someone asks a follow-up question, but a specific, checkable account of which sections, and how much.
The limits are the same ones that apply everywhere else in this system. It doesn't establish creative or expressive judgment on its own: typing text isn't the same as exercising the "sufficient expressive elements" standard the Office actually applies. It only covers text written inside the editor, so anything drafted elsewhere and pasted in shows as a paste, not a traced origin. What it does is turn "describe the human contribution" from a reconstruction exercise into something you can just look up.