AI and Copyright for Authors: What the US Copyright Office Has Said

The US Copyright Office's position, in its January 2025 report, is that copyright protects the human-authored expression in a work even if the work also contains AI-generated material, but not purely AI-generated material, and that, based on current generally available technology, prompts alone do not provide sufficient control. That is the Office's analysis of existing US law, not a court ruling, and it applies to the United States. This is general information, not legal advice. If you have a specific situation, such as a dispute, a contract or a registration problem, talk to a lawyer.

Copyright law and KDP's disclosure rule are different things

Whether you can copyright a book and whether you must tell a retailer about AI are separate questions with separate sources. Copyright is a matter of law and of the Copyright Office's practice. KDP's AI disclosure is a condition of using KDP, a contract-style platform rule that applies whatever copyright law says. A book can be fully protected and still require a KDP disclosure, and the reverse is also possible. The platform side is on the KDP AI content disclosure page.

What we read, and how

We extracted the text of two Copyright Office PDFs on November 8, 2026 and read the passages cited below, not every page of Part 2. The PDF text did not come through our web reader, so we worked from the downloaded files. We did not read Part 1 (digital replicas, July 31, 2024) or the pre-publication Part 3 on training (May 9, 2025). The dates and the list of parts come from the Office's AI index page, which does not itself describe the guidance's content beyond its title.

Part 2 of the report: copyrightability (January 2025)

The Part 2 report is dated January 2025, and the index page gives January 29, 2025. Its executive summary lists these conclusions, which we paraphrase:

• Existing law is enough, and legislative change is not needed.
• Using AI tools to assist rather than stand in for human creativity does not affect whether the output can be protected.
• Copyright protects the original expression in a work made by a human author, even if the work also includes AI-generated material.
• Copyright does not extend to purely AI-generated material, or to material where there is insufficient human control over the expressive elements.
• Whether human contributions are enough to count as authorship is decided case by case.
• "Based on the functioning of current generally available technology, prompts do not alone provide sufficient control."
• Human authors are entitled to copyright in their works that are perceptible in AI-generated outputs, as well as in the creative selection, coordination or arrangement of material in the outputs, or creative modifications of the outputs.
• The Office said the case had not been made for extra copyright or sui generis protection for AI-generated content.

The report also says it will keep monitoring developments and may revisit these points, and that its conclusion that prompts alone are unlikely to be enough is tied to current technology. The report is written as general analysis across all kinds of works, and a search of its text finds no discussion of novels or audiobooks as such.

How the report reasons about each kind of use

The body of the report goes further than its summary. These five passages are the ones an author is most likely to need. We paraphrase, and page numbers are the report's.

Assistive use (pages 11 to 12). The Office draws "an important distinction between using AI as a tool to assist in the creation of works and using AI as a stand-in for human creativity". It says the distinction "depends on how the system is being used, not on its inherent characteristics".

Brainstorming and outlines (page 12). The report notes a commenter's point that AI can be used to structure or create a preliminary outline for literary works. Where the user is "referencing, but not incorporating, the output" in developing their own work, the Office says using AI that way should not affect the copyrightability of the resulting human-authored work.

Prompts and re-prompting (pages 18 to 20). The Office says prompts function as instructions that convey unprotectible ideas. Even detailed ones do not, at present, control how the system generates the output. Revising a prompt many times does not change this: it compares re-prompting to re-rolling the dice, and says time and effort are irrelevant, because copyright protects original authorship and not hard work.

Putting your own work in (pages 22 to 24). The report separates prompts that merely describe an outcome from inputs where human expression is retained in the output. A commenter's example was uploading a first-person story and asking for it in the third person. The Office's conclusion, drawn from a hand-drawn image case, is that where a human inputs their own copyrightable work and that work is perceptible in the output, they will be the author of at least that portion. The protection covers the perceptible human expression and does not extend to the AI-generated elements alone. The report's own example is visual, and it contrasts a refused case where the input was not clearly visible in the output.

Editing and arranging AI material (pages 24 to 27). Humans may select or arrange AI-generated material in a sufficiently creative way, or modify it enough to meet the standard for copyright. That protects the human contribution and the work as a whole, not the AI-generated content alone. The report calls human text combined with AI images "a relatively common scenario" in registration applications, and describes a comic book where the selection and arrangement of images with human text was protected as a compilation. A footnote adds that arranging only two or three elements is generally not enough. Including AI-generated elements in a larger human-authored work does not affect the copyrightability of the larger work as a whole.

What a court has said, as the report describes it

The report says that in 2023 the US District Court for the District of Columbia became the first court to address the copyrightability of AI-generated outputs specifically, in Thaler v. Perlmutter. As the report describes it, the court affirmed the Office's refusal to register an image described as created autonomously by an algorithm, and said "human authorship is a bedrock requirement of copyright". The report notes the decision had been appealed. We did not read the decision or any later ruling, so we do not say how the appeal ended.

The March 2023 registration guidance

The registration guidance was published March 16, 2023 in the Federal Register and is effective that date. It deals with what an applicant must say when registering a work that contains AI-generated material. As read:

• Applicants have a duty to disclose the inclusion of AI-generated content in a work submitted for registration, with a brief explanation of the human author's contributions.
• You claim your own contributions, using the Standard Application, and describe the human authorship in the "Author Created" field.
• AI-generated content that is "more than de minimis" should be explicitly excluded from the claim.
• Applicants should not list an AI tool or its company as an author just because they used it.
• Applicants who are unsure can give a general statement that the work contains AI-generated material, and the Office will follow up.

This is guidance for registering. It does not say that unregistered works are unprotected, and we did not read the Office's other registration decisions listed on its index page.

Filling in an application

The guidance names the fields, which makes the process more concrete than most summaries suggest.

If you have already applied or registered without saying

The guidance covers this too. For a pending application, contact the Office's Public Information Office and report the omission. For a completed registration, submit a supplementary registration that describes the human contribution and disclaims the AI material. It warns that applicants who do not update the record "risk losing the benefits of the registration". The Office may take steps to cancel the registration, and a court may disregard a registration under section 411(b) of the Copyright Act if the applicant knowingly gave inaccurate information that would have led to refusal.

What this means for different kinds of AI use

The sources support a few careful points, and no more. This table is our reading of the two documents. Neither one rules on these exact facts, and the Office says these questions are decided case by case.

Compare this with KDP. An AI draft you rewrite heavily may contain protectable human authorship, and still be AI-generated under KDP's wording. For drafting habits that keep your own voice in the work, see using AI without losing your voice. For the same questions applied to images, see AI cover art for authors.

Practical records

This is general advice, and the documents do not require it. Because the Office asks for a description of the human contribution and decides cases on how a tool was used, a dated record helps. Keep early drafts, note which passages came from a tool, and keep versions that show your changes. If you may register, write your "Author Created" description while the facts are fresh.

What we did not cover

This page covers only the US Copyright Office documents above. We did not read courts' decisions, other countries' law (the report has a section on the international landscape that we did not summarize), AI training litigation, or the terms of any AI tool about who owns output. A tool's terms may say what its vendor claims or assigns, but they are a different question from whether copyright exists. Retailer labelling for audio is in the audiobook AI labelling guide. For the tools an author might choose, see the self-publishing software stack. The AI policy and law hub lists the rest.

Last verified 2026-11-08. Rules and prices change: check the linked primary page before you rely on this.