What the Copyright Office Actually Said, and What People Think It Said

The Actual Ruling

In February 2023, the US Copyright Office issued a registration decision for Zarya of the Dawn, a graphic novel created by Kristina Kashtanova using Midjourney-generated images alongside her own text and visual arrangement. The Office's conclusion was precise, and it has been consistently misread in the two years since.

The Office did not refuse to register the work. It issued a revised certificate covering the text, the selection, and the arrangement — every element Kashtanova had authored in a legally recognisable sense. What it refused was copyright protection for the individual Midjourney-generated images themselves. Those images, the Office found, were not the product of human authorship. Kashtanova had provided text prompts, but the Office determined that prompting a generative system does not constitute the kind of creative control that copyright law requires. The machine, not the human, made the expressive choices that produced each image.

The Office grounded this in the foundational doctrine that copyright subsists in works of human authorship. That principle predates digital imagery entirely; courts have applied it to disqualify photographs taken autonomously by wildlife cameras and, in the Naruto monkey-selfie litigation, images taken by a macaque. What changed in the Kashtanova decision was the application of that doctrine to a generative AI workflow for the first time as formal Office guidance.

What the Guidance Extended

In March 2023, the Office published broader registration guidance on works containing AI-generated material, elaborating the framework. The position the Office articulated is that the copyrightability of AI-assisted work is determined element by element. Human-authored text, human-made selections, human arrangements of generated material: potentially protectable. The generated image itself, absent sufficient human control over its specific visual expression: not protectable.

The Office has been careful to avoid drawing a fixed line at the prompt. It acknowledged that the quantity and nature of human input matters and that future workflows might involve degrees of control that change the analysis. What it declined to do is declare prompting, as currently practised in systems like Midjourney, sufficient authorship. The system's internal processes — how it selects, weights, and renders — are not transparent to or directed by the user in the granular way that operating a camera, even a highly automated one, is.

That camera comparison is not incidental. Photographers routinely rely on automated metering, autofocus, and computational processing. Copyright has never required that every exposure decision be manual. What it has required is that a human being stand as the originating creative intelligence. The Office's position is that a prompt does not satisfy that requirement when the expressive result is substantially determined by the model.

The Misreading and Its Consequences

The popular summary — "you can't copyright AI art" — is accurate as far as it goes and misleading in how it tends to be deployed. It is used to argue both that all AI-assisted work is unprotectable (wrong: a human author who meaningfully selects and arranges generated elements retains rights in that arrangement) and that the Office has somehow resolved the broader question of AI training and copyright (it has not; that question involves different doctrine entirely and remains actively litigated).

For working photographers, the practical stakes are cleaner. Images produced by generative systems without meaningful human authorial control enter the public domain immediately upon creation — no registration, no term, no exclusivity. The market implications of a large and growing body of freely usable imagery are significant, and they flow directly from a decision most people have only encountered secondhand.