Mostly no, not from a prompt alone. To copyright AI-generated art in the United States, a person has to have authored it, because copyright protects what a human author creates, and an image a generator produced from your text instructions is not protected by itself. What you can protect is the human part: original artwork you feed in, creative edits you make, and the way you select and arrange pieces.

That has been the Copyright Office’s position since 2023, and in March 2026 the Supreme Court declined to disturb it. Below is a plain reading of what is settled, what is not, and what to do if you want a registration. This covers US federal law; other countries differ.

The short answer, sorted by situation

What you did Likely protection in the US
Typed a prompt, kept the image as generated None for the image itself
Wrote many prompts and picked the best result Still none for the image, according to the Copyright Office
Fed in your own drawing or photo and kept it visible in the result Your original input, plus possibly the way it is combined
Edited the output by hand, repainting or compositing The human changes, if they are creative enough
Arranged AI images with your own text into a book or layout The selection and arrangement, and your text
Named a machine as the author Registration refused

Myth versus fact

Myth: AI art can never be copyrighted. Fact: work made with AI help can be registered when a person contributed enough original expression. The D.C. Circuit said so itself in the Thaler case, noting that its ruling does not bar works made with the assistance of AI; it only rules out listing an AI system as the author.

Myth: a long, detailed prompt makes you the artist. Fact: the Copyright Office’s January 2025 report concluded that, with today’s tools, prompts function as instructions, and the system decides how they become visible expression. Rewriting and resubmitting the prompt did not change that analysis, nor did the effort involved. The Office has likened a prompt to directions given to a commissioned artist, who, not the client, is the author.

Myth: if the Supreme Court did not take the case, the question is closed. Fact: refusing review settles nothing on the merits. It leaves the D.C. Circuit’s ruling in place without creating a national rule, and the harder question, how much human input is enough, was never reached.

What happened in Thaler v. Perlmutter

Stephen Thaler applied to register an image titled “A Recent Entrance to Paradise.” He said it was created autonomously by his system, which he called the Creativity Machine, and he named the machine as the author. The Copyright Office refused, a district court agreed, and the D.C. Circuit affirmed on March 18, 2025.

The appeals court read the Copyright Act as assuming a human author. Ownership, a copyright term measured by an author’s lifespan, and the signature on a registration all point that way. It also treated the Office’s long-held practice as something Congress carried into the 1976 Act. A work-made-for-hire theory failed for the same reason, and a late argument that Thaler himself was the author by virtue of building the machine was waived because he had not raised it earlier.

Thaler petitioned the Supreme Court in October 2025. The Solicitor General urged the justices to say no, partly because he had disclaimed any human creative contribution, which made the case a poor test of the harder question. The Court denied review on March 2, 2026. Because he had told the Office there was no human contribution at all, the case never forced a ruling on any middle ground.

A short timeline of how we got here

Date Event
2019 Thaler applies to register “A Recent Entrance to Paradise” with the machine named as author
March 16, 2023 Copyright Office registration guidance for works containing AI-generated material takes effect
September 2024 Jason Allen sues the Copyright Office in Colorado over the refusal for his Midjourney image
January 29, 2025 Office publishes Part 2 of its AI report, on copyrightability
March 18, 2025 D.C. Circuit affirms in Thaler v. Perlmutter
October 2025 Thaler asks the Supreme Court to hear the case
March 2, 2026 Supreme Court denies review

The date for Allen’s filing comes from the docket entry showing his complaint was filed on September 26, 2024. Everything after that, as far as I can verify, has been briefing rather than a decision.

The arguments the Supreme Court never had to resolve

Thaler’s petition made a bolder claim than most people assume. He argued that the human-authorship requirement is an invention of the agency, with no basis in the statute or the Constitution. He also pointed to work-made-for-hire, to the Supreme Court’s Star Athletica decision, and to the way photography was once treated, and warned that denying protection would put American AI investment at a disadvantage against countries such as China and the United Kingdom.

The government’s reply was narrow on purpose. It said the case asked only whether a machine can be an author, that the D.C. Circuit read the word correctly, and that the Office had by then registered hundreds of works containing AI material where a person’s contribution was shown. A petitioner who insists no human contributed anything is, from the government’s side of the table, the easiest case to defend.

Thaler’s reply brief tried to widen the frame. He argued the Office’s practice amounts to a flat rule against AI-assisted works, citing refusals even where applicants described extensive prompting. That tension is real, and it is exactly why the next case matters more than this one. A court that sees a record of real human creative steps, rather than an applicant who disclaims them, may be asked to say where a line falls.

Where the line is drawn: three routes to protection

The Office’s guidance, first issued as a policy statement effective March 16, 2023 and expanded in its January 2025 report, gives three routes.

1. Your own expression, still visible

If you supply original material, such as a sketch, a photograph or a written passage, and it remains perceptible in the output, you are the author of that portion. The Office treats the protection like a derivative work: it covers what you contributed, not the generated surroundings. Its report contrasted two decisions here. A work called Rose Enigma was registered with the claim limited to the human-drawn elements that remained perceptible, while Suryast was refused because the applicant’s control over what the AI produced was not enough and the input was not clearly visible in the result.

2. Creative modification

Repainting, retouching with real artistic choices, compositing, and similar changes can count as new authorship. Whether a given edit is original enough is judged case by case. Ordinary tools such as editing software do not remove human authorship, though AI-powered features inside them are examined individually.

3. Selection and arrangement

The best-known example is Zarya of the Dawn, a comic. The Office protected the text and the selection and arrangement of the elements, but not the individual Midjourney images. If you assemble a collection or a layout with real creative choices, that assembly can be yours even where each piece is not.

How the rules play out for common creators

Abstract rules are easier to follow with examples. These are illustrations of the Office’s published reasoning, not predictions about any particular application.

A hobbyist who posts generated images online

Nothing in the law stops you from posting them, and nothing in US copyright law will stop a stranger from copying an image that is entirely machine-made. Platforms may have their own rules, and the tool’s terms of service may say you own the output as between you and the company. That is a contract matter. It is not a copyright, and it does not bind anyone outside the agreement.

A designer who generates a rough, then repaints it

This is the strongest pattern for registration. The final piece shows your own brushwork or compositing, you can document the steps, and you can truthfully describe which elements are yours and which came from the generator. Expect to disclaim the portions that are straight output.

A writer who uses AI for brainstorming or outlines

The Office’s report said that using AI as an assistive tool, for something like song ideation or a literary outline, does not by itself affect copyrightability of the finished human work. The line is crossed when the machine, rather than the person, writes the expression that ends up in the final text.

A publisher assembling a picture book

Here the Zarya of the Dawn example applies most closely. The words you wrote, and your decisions about order, pairing and layout, can be protected. Each generated illustration, standing alone, cannot. If another publisher lifts a single page, your protection for that one image may be thin or absent, while copying the book’s layout and text would still raise a claim.

A trademark is a separate tool again. A logo or character name can function as a mark based on use in commerce, without anyone being its copyright author, though that is a different body of law with its own requirements, and a generated logo that resembles an existing mark can create trouble of its own.

What is still open

Three questions remain unanswered by any court I could find.

  • How much input is enough. Commentators say not to expect a percentage. Courts and examiners will weigh what you can show you contributed.
  • Heavily prompted images. Jason Allen made “Théâtre D’opéra Spatial” with Midjourney using a very large number of iterative prompts and sued the Copyright Office in Colorado in 2024 after it refused registration. Sources I found describe the case as pending; I could not confirm a ruling as of this writing.
  • Congress. The Copyright Act never uses the words “human author,” and Congress has not amended it. The Copyright Office said in its report that it saw no need for new legislation, so a change looks unlikely soon.

The Office also said its conclusion depends on current technology. If tools someday give users direct control over specific expressive elements, the analysis could shift.

Why this rule exists, and why it may not last forever

The Office gave several reasons for declining to protect pure machine output. It said AI developers already have strong incentives to build their products, so copyright is not needed to reward them. It worried that a flood of protected machine-made material could crowd out or devalue human creative work. And it did not accept international competition as a reason to change course.

It also left the door ajar. Its conclusion rests on how today’s generally available tools behave: the user supplies instructions, and the model makes the expressive choices. If a future tool let a person dictate specific expressive elements directly, the answer for that tool could differ. Automated prompt optimization, the report added, could push the other way by reducing user control further. In other words, the law is not claiming that machines can never be tools of authorship, only that the typing of a prompt has not yet been enough.

If you want to register a work that used AI

  1. Keep records. Save your drafts, your inputs, and notes on which parts you drew, edited or arranged. You may need to describe your contribution.
  2. Disclose. Applicants must report AI-generated content that is more than minimal and briefly describe the human author’s work. The Standard Application is the route the Office points to.
  3. Claim only your part. Describe your contribution in the author-created field and list the generated material as excluded. Never name an AI tool or its maker as an author.
  4. Fix mistakes. If you registered earlier without disclosing AI use, the Office describes a supplementary registration. Leaving an inaccurate registration alone can lead to cancellation, and a court may disregard it in an infringement case if you knowingly gave wrong information.

Questions an examiner will effectively ask you

The Office does not publish a scoring rubric, and nothing here guarantees an outcome. Still, its guidance and report imply a set of questions worth answering before you file.

  • Which parts of this work did I make myself, and can I show them?
  • Where the AI generated something, how much control did I actually have over its specific expressive details, as opposed to the general idea?
  • Did I choose, order or combine elements in a way that required creative judgment?
  • Is my own input still visible and recognisable in the final result?
  • Have I described the AI-generated material accurately, rather than hoping the examiner will not ask?

The last question carries real weight. The Office’s guidance warns that a knowingly inaccurate application can lead to cancellation or to a court disregarding the registration. Candor is cheaper than a lawsuit over an invalid certificate.

What this means in practice

An unprotected image can still be used, and so can copies of it. Anyone may generally reuse the unprotected portions without permission, which matters if you sell prints, license artwork, or build a brand around generated visuals. Contract terms may give you rights as against the tool’s company, but they cannot create a copyright the law does not grant. That is the subject of a companion question about who owns what an AI tool wrote for you.

For work that matters commercially, the safer pattern is to treat the AI output as raw material and put real, documentable human authorship on top of it. Then you have something you can describe to an examiner, and something a court can recognise.

Related questions

Can I copyright a book I wrote with ChatGPT? The human-written text is protectable; passages generated verbatim by the tool generally are not. The same selection-and-arrangement logic applies, as in the Zarya of the Dawn registration.

Does the Copyright Office refuse every AI-assisted work? No. The Office says it has registered hundreds of works containing AI-generated material, limited to the human contribution.

Do other countries agree? Not uniformly, and this post covers only the United States. The Copyright Office said rules emerging abroad were generally consistent with its approach, but check local law if you work internationally.

This is general information, not legal advice.

Sources