If You're Using AI to Make Art, the Supreme Court Just Sent You a Message
After years of litigation, courts are clarifying what role humans must play for AI-assisted work to qualify for copyright.
So you’ve finally completed your Rauschenberg-meets-O’Keeffe series of risqué décolletage generative AI paintings, and you think it’s time to make things official before you start contacting galleries and competitions. OR you just finished your brilliant ode to early Palahniuk-meets-Mamet literary novel, mostly written by AI, of course. And now you’re wondering what happens when you send it in to the U.S. government to be copyright-protected so that it can later be optioned and turned into a groundbreaking 8-episode prestige TV miniseries on Netflix.
Well, not so fast, my tech-enabled wordsmith. There’s something you need to know about your AI creative portfolio, and it could change everything about how you use AI to wrap yourself in the algorithmic cloak of aesthetic prestidigitation.
On Monday, the U.S. Supreme Court denied [PDF] a motion to hear an appeal from computer scientist Stephen Thaler, who has been attempting to get his AI artwork recognized as a copyrighted work since 2018. The U.S. Copyright Office (USCO) rejected his initial copyright registration attempt in 2019. Then the USCO denied his request for reconsideration in 2020 and again in 2022.
That’s when Thaler decided to file a lawsuit in the U.S. District Court in hopes of having his AI images made eligible for copyright. In 2023, a federal judge upheld the decision of the USCO. Later, the U.S. Court of Appeals for the D.C. Circuit upheld the lower court’s decision in 2025. So Thaler decided to take his case to the highest court, the Supreme Court, in October 2025.
And that brings us to this week, when the U.S. Supreme Court declined to hear Thaler’s appeal, thus upholding the previous decisions in the lower courts.
The legal arguments contained in the many legal filings are serpentine and exhaustive in rebuking Thaler’s claim, but the overall gist is: No, typing prompts into an AI image generator doesn’t count as “authorship” worthy of copyright, according to the USCO and the courts that upheld its decision.
Spurred on by this case and the growing waves of AI users with similar copyright concerns, in January 2025, the USCO published guidelines to help AI users [PDF] understand what is and isn’t copyrightable. What we now know is that you can use AI to create copyrightable works, but those works must contain significant human input and/or modification, and only those human-made aspects are generally eligible for copyright.
A human may “modify material originally generated by AI technology to such a degree that the modifications meet the standard for copyright protection,” reads the USCO document. “Similarly, the inclusion of elements of AI-generated content in a larger human-authored work does not affect the copyrightability of the larger human-authored work as a whole. For example, a film that includes AI-generated special effects or background artwork is copyrightable, even if the AI effects and artwork separately are not.”
This is all good news, that is, assuming you’re a creative individual who likes to make things the traditional way, and maybe are just looking for a bit of AI seasoning. But if you’re among those who simply want to type a prompt and suddenly claim to be an artiste, with all the concomitant privileges and plaudits conferred by such trappings, well, the USCO doesn’t really see you.
The Authorship Test
The root of some of the current copyright logic comes from a Supreme Court case back in 1991, Feist Publications, Inc. v. Rural Telephone Service Co., which led to a distinction being made between supposed “sweat of the brow” facilitated by manipulating data or mechanisms, versus truly novel creative output from the human. How that novel creative output is determined is generally down to the matter of control. Do you have it? Or are you relying on a mechanism to churn out product that may or may not represent a unique human expression?
This issue is addressed in the 2025 USCO AI guidelines. “[AI] Prompts alone do not provide sufficient human control to make users of an AI system the authors of the output … While highly detailed prompts could contain the user’s desired expressive elements, at present they do not control how the AI system processes them in generating the output,” reads the USCO guideline paper. “Copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements.”
So you can use AI to write that book, create that graphic, or direct that video, but if you’re hoping to let the AI do 90% of the heavy lifting and “creative” knitting, the USCO won’t be able to protect you if someone comes along and rips off your AI-generated work for their own purposes. The laws may evolve, and the USCO is making its AI vs. human copyright decisions case-by-case, but this is, in very broad strokes, where we’re at right now.
AI artist Kristina Kashtanova got an early taste of this handling back in 2023. She registered her AI-generated comic book with the USCO successfully, but without properly disclosing that she used Midjourney to create the images. When word got back to the USCO, the office revoked the copyright on the overall comic book. The office then issued a copyright [PDF] that only applies to the text “as well as the selection, coordination, and arrangement of the Work’s written and visual elements.” But the images, the lifeblood of any good comic book, are not copyrighted.
The Detection Gap
“Applicants have a duty to disclose the inclusion of AI-generated content in a work submitted for registration and to provide a brief explanation of the human author’s contributions to the work,” a 2023 policy memo from the USCO reads.
Kashtanova could have avoided her high-profile copyright hassles if she had, as Thaler did, properly disclosed that AI-generated imagery was central to the work she was attempting to copyright. In her case, the USCO apparently heard about the AI component of the work via the media. But what happens if the person seeking the copyright doesn’t disclose that the work is all or partially AI-generated?
Surely that means the USCO has a rigorous backend methodology designed to examine and determine when AI-generated work is sent in that may run afoul of its guidelines. Right? Well, so far, the answer to that appears to be NO.
Right now, it appears that the USCO is operating on an honor system, leaving disclosure of the nature of a work sent to its office up to the copyright applicant. And for a relatively good reason. AI detectors are not yet reliable enough to consistently—in a manner that stands up to peer-reviewed research—detect what is and isn’t AI writing or AI imagery. If you’re like me and you’ve used these AI detection tools for a few years, you’re probably pretty confident in their efficacy. I know I am. But that AI detection ruling must be able to stand up in court. Today’s AI detectors still aren’t quite there. So until AI detectors meet the Daubert standard of scientific evidence, the USCO won’t have the tools it needs to render such judgments on what is and isn’t AI when it comes to copyright candidates.
This means there’s likely a decent amount of material, both written and visual, currently registered with copyrights at the USCO that are fundamentally invalid, it’s just that the court-ready AI detection tools haven’t arrived yet. But when they do, USCO will have their work cut out for them, and copyright attorneys could enjoy a bonanza of plaintiff and defendant fees as undisclosed erroneously copyrighted AI-generated creations are exposed as being, well, unprotectable, and thus devoid of inherent value.
Of course, it’s important to remember that this only applies to works that human recontextualizations and/or additions haven’t significantly modified. So if the AI-using creator isn’t lazy and does add their own transformative touch to the AI-generated work, this copyright vulnerability may be less of a concern. But for the rest of them, leaning perhaps too heavily on AI to stand in for their less-than-perfect human creativity muscles, the AI honeymoon will be at least interrupted, and plans for a post-nuptial agreement with the public will need to be hastily drawn up. ✍︎
New essays on AI, film, & entertainment innovation arrive by email. Subscribe to stay in the loop. All editorial text is written by humans.
One important takeaway here for creators is not panic. It is process.
If you are creating original work, keep your original documents. Keep your drafts. Keep your revision history. Keep separate saved versions that show how the piece developed over time.
That is not just good discipline anymore. That is evidence.
In a world where people can casually accuse anyone of having AI write for them, creators need a clean paper trail that shows where the work began, how it changed, and what human judgment shaped the final result. Prompting may be cheap. Authorship should not be.