A Recent Entrance to Paradise is a painting that frequently appears in legal circles. Its color scheme is dreamy, lush, and almost romantic. The kind of piece you might find in a gallery in lower Manhattan, with the artist’s name and the year written on a little white card next to it. However, there isn’t an artist. Computer scientist Stephen Thaler’s algorithm, DABUS, produced the entire piece.
Thaler applied for copyright protection but was categorically turned down. Upheld by the courts and upheld by the Supreme Court in March 2026, that denial has quietly emerged as one of the most significant legal developments in recent creative history.
| Category | Details |
|---|---|
| Topic | AI-Generated Art & Copyright Law |
| Primary Legal Body | U.S. Copyright Office |
| Landmark Report | Copyright and Artificial Intelligence: Copyrightability (January 2025) |
| Key Legal Case | Thaler v. Vidal — Stephen Thaler & DABUS AI System |
| Supreme Court Action | Declined to hear AI copyright case (March 2026) |
| Core Legal Principle | Human authorship required for copyright protection |
| Key AI Platforms Involved | Midjourney, DALL·E, Stable Diffusion, RAGHAV |
| Notable Registered Works | Zarya of the Dawn, A Collection of Objects Which Do Not Exist, Just Like In A Movie |
| Next Report Expected | Part 3 of U.S. Copyright Office AI Report — Late 2025 |
| Reference Website | U.S. Copyright Office — AI and Copyright |
Can you copyright art created by a machine? That seems like a straightforward question. In reality, responding to it has drawn lawyers, artists, legislators, and tech firms into a protracted dispute for which no one appears to be completely ready.
In January 2025, the U.S. Copyright Office published Part 2 of its historic report on AI and copyright, confirming what courts had been implying for years: human authorship is a prerequisite for copyright protection. Artificial intelligence-only works that lack significant human creative input are ineligible. The line is firm. However, it turns out that the creative world seldom follows a linear path.
Nowadays, you can find artists using Midjourney or DALL·E as a tool, not as a substitute for Photoshop, in any major design studio. They spend hours editing and writing the final piece after drafting prompts, refining outputs, and discarding dozens of iterations before choosing one. To its credit, the Copyright Office has recognized this fact.
The 2025 report makes a careful distinction between a creator who exercises consistent creative judgment throughout the process, shaping and editing AI-generated elements into something that reflects their own vision, and a person who types a simple prompt and accepts whatever the machine produces, which does not qualify for protection. Even though it’s still quite challenging to pinpoint the precise boundary, that distinction is crucial.
This tension was early and evident in the case of Zarya of the Dawn. Midjourney was used by Kristina Kashtanova to create a graphic novel, which was first registered with the Copyright Office. The Office ultimately only granted protection to the text and the human-authored selection and arrangement of elements after canceling a portion of the registration after learning that the images were AI-generated. Despite their striking quality, the individual images were not protected.
Although Kashtanova had produced more creative work than most people realized from the outside, it’s possible that she was disappointed by this result. According to the ruling, if the machine is making the final expressive decisions, even intense creative engagement with an AI tool might not be sufficient.
The CEO of the AI platform Invoke, Kent Keirsey, recently registered an image titled “A Single Piece of American Cheese,” which only makes sense when you consider the dry wit that permeates the entire business. Keirsey contended that his practical application of an inpainting technique, which allowed him to alter and improve 35 different aspects of the picture, constituted adequate creative input.
The Copyright Office concurred. That choice is important because it starts to define what “enough human involvement” actually means in practice, not because of how the image looks. 35 intentional changes, meticulously recorded. It establishes a peculiar but oddly helpful precedent.
The same obstacles are encountered by software developers. For the parts of IBM’s Watsonx Code Assistant that were written by humans, copyright registration was granted; AI-generated code was specifically left out. The same was true for Adobe’s GenStudio for Performance Marketing, where GitHub Copilot’s contributions were completely excluded and protections were only given when a human hand was obviously at work.
The Copyright Office appears to be enforcing what legal observers refer to as a “bright-line rule,” which states that regardless of how well AI is incorporated into the finished product, it is not covered if it was created.
This creates a great deal of uncertainty for companies that have built entire workflows around AI-assisted content generation. Every time a machine completes a significant amount of the work, publishers, game developers, and marketing companies now have to deal with the difficult question of whether their outputs can truly be protected or if they are essentially publishing into the public domain.
Even though the legal reasoning is sound, the wider implications are a little unsettling. The existing framework permits anyone to extract and reuse those unprotected elements without facing any repercussions by placing AI-generated content in the public domain. Theoretically, a business could recognize the parts of a competitor’s hybrid work that were created by AI and freely replicate them, as long as they refrained from replicating the parts that were specifically written by humans.
Depending on who you ask, that could be a feature of the system or a loophole. It’s still unclear if courts will progressively create doctrines that make it more difficult to dissect hybrid works or if future legislation will close that gap.
All of this lacks a definitive response to the issue of collaborative ownership, which the Copyright Office’s 2025 report subtly acknowledges but does not fully resolve. Who owns the final product if an AI tool improves an artist’s rough sketches—the artist or the business that licensed the AI? How do you even start mapping the authorship of a single creative work when multiple humans and multiple AI systems contribute to it? These are not speculative inquiries.
Real contracts, real disputes, and real studios are already experiencing them. The topic of AI training data and licensing is expected to be covered in Part 3 of the Copyright Office report, which is anticipated to be released sometime in late 2025. This topic is expected to cause even more controversy than the authorship issue.
As this develops, it seems as though the legal system is making every effort to apply century-old principles to a situation for which it was never intended. It is not incorrect for courts to demand human authorship. This idea has a long history, dating back at least to Burrow-Giles v. Sarony in 1884, when the Supreme Court affirmed copyright for photographs because a human had made intentional, artistic decisions regarding composition, lighting, and posing. The reasoning from then is still valid today.
When a machine runs patterns from a dataset, it computes probabilities rather than making any kind of meaningful decisions. However, the artists who are directing those computations, creating prompts over hundreds of iterations, choosing, organizing, and editing the outcomes, and making aesthetic choices at every stage—they are doing something. Courts and regulators will be debating the precise amount that is sufficient for years to come.
For the time being, the practical advice for artists is straightforward, if a little annoying: record everything, make deliberate changes, and never let the machine make the final decision. As it stands, humans own copyright. The machine is able to assist. It simply cannot possess anything.
