AI-Assisted Writing and Copyright: What Authors Need to Know Before Submitting or Publishing

The Human Authorship Requirement and What It Means for Writers Using AI

Copyright protection in the United States has always required a human author. That principle predates the digital age by more than a century. In Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884), the Supreme Court grounded copyright in "original intellectual conceptions" of a human mind. The rule has never changed. AI has only made it urgent.

In March 2025, the D.C. Circuit confirmed the modern application of that rule. In Thaler v. Perlmutter, the court held that "as a matter of statutory law, the Copyright Act requires all work to be authored in the first instance by a human being." An AI system cannot be an author. Work generated entirely by AI — with no human creative contribution — receives no copyright protection.

That holding does not, however, foreclose copyright protection for work that involves genuine human authorship alongside AI tools. The court's decision addressed a work generated entirely by AI with no human creative contribution — a fundamentally different scenario from an author who uses AI as part of a creative process. The Copyright Office's Part 2 Report addresses what protection is available when a human author does use AI: the question is how much human creative input is present, and where in the process it appears.

The U.S. Copyright Office addressed that question directly in its Part 2 Copyrightability Report (January 2025). The standard it articulated: "sufficient human control over the expressive elements" of the work. Prompts alone do not satisfy this standard. Generating text from a prompt and publishing it as written gives you no copyright in the AI's output. What does qualify — according to the Office — is creative selection, arrangement, or coordination of AI-generated material, or using AI output as reference that you then transform through your own writing.

For practical purposes, the distinction that matters is between AI as a generator and AI as a collaborator. If you used an AI tool to brainstorm, outline, or draft sections that you then rewrote, reorganized, or selected from based on your own literary judgment, your human authorship is traceable. If the published text is substantially what the model produced with minimal intervention, your copyright claim is weak regardless of how the work is labeled.

How to Register AI-Assisted Work — and What to Disclose to the Copyright Office

The Copyright Office's March 2023 guidance, published at 88 Fed. Reg. 16190, gives writers a clear filing path for AI-assisted work. Use the Standard Application. In the "Author Created" field, describe only the human-authored portions of the work — the plot structure you built, the prose you drafted, the narrative voice you shaped. In the "Material Excluded / Other" field, disclaim the AI-generated content with language such as "[description of content] generated by artificial intelligence." If you added new human-authored material to AI output, complete the "New Material Added / Other" field as well. An optional "Note to CO" field exists for context that doesn't fit elsewhere.

There is one narrow exception to the disclosure requirement. If AI use is de minimis — meaning the AI-generated material, standing alone, would not satisfy the originality threshold the Supreme Court set in Feist Publications v. Rural Telephone Service — no disclaimer is required. In practice, a single AI-suggested sentence reworked into your prose is unlikely to cross that line. An AI-generated chapter is a different matter.

The stakes of getting this wrong are concrete. In Zarya of the Dawn, the Copyright Office cancelled the original registration that covered AI-generated images and issued a narrower certificate covering only the human-authored text and the human-authored selection and arrangement of the compilation. The underlying lesson is not subtle: registration that overclaims human authorship is vulnerable. The Authors Guild AI Best Practices go further, warning that knowingly failing to disclose AI-generated text may be deemed fraud on the Copyright Office and invalidate the registration entirely.

One more scenario authors miss: if your publisher is filing the registration on your behalf, you are still responsible for the accuracy of the application. The Authors Guild advises that authors must inform their publisher of any AI use so the application reflects the work's true authorship. The obligation to disclose does not transfer with the filing responsibility.

What Major Publishers and Literary Agents Require Now

None of the Big Five publishers have issued a house-wide rule forbidding authors from using AI during composition. Their existing contract language targets a different problem: preventing third parties from using the publisher's catalog to train AI models. Penguin Random House, Hachette, and Macmillan have added anti-training clauses to new contracts, with PRH going as far as printing "no AI training is permitted" on copyright pages — but these provisions protect the publisher's intellectual property from scraping, not the author's writing process from AI assistance.

That distinction matters beyond an author's own practices. AI exposure can enter a manuscript through collaborators — a freelance editor, developmental editor, or writing contractor who uses AI tools without the author's knowledge or disclosure. The risk from third-party AI use is the same as direct use: if undisclosed AI-generated text appears in the submitted manuscript, the author bears the contractual and legal consequences, not the collaborator. That is why AI disclosure obligations in your contracts with collaborators matter as much as AI clauses in your contracts with publishers.

Literary agents are moving faster than publishers. Agencies have begun implementing submission management tools to screen for disclosed AI use, and AI detection software is under active evaluation across the industry. Query forms are beginning to include the question directly: "Was any part of this book or query package created by AI?" Disclosure — voluntary or compelled — is now a submission variable, not an afterthought.

Academic and nonfiction publishers have already moved from practice to policy. Taylor & Francis mandates disclosure of any generative AI use, requiring authors to name the tool, version, and purpose, with placement in the Methods or Acknowledgments section for articles and in the proposal or introduction for books. Elsevier, Springer Nature, Wiley, Taylor & Francis, and SAGE all prohibit listing a generative AI system as an author or co-author — a rule that sounds obvious but reflects how many early submissions attempted to attribute work to AI.

The commercial pressure on publishers to formalize these rules is intensifying. Author organizations pressed the Big Five through 2025 to stop acquiring AI-generated books and halt internal AI replacements of editorial staff, drawing significant support from writing communities within days of each petition launch. Industry policy is reactive to author pressure, litigation risk, and reputational exposure — which means the informal norms in place today will likely become explicit contract terms within the next publishing cycle.

Contract Clauses to Watch For — and What the Authors Guild Recommends

Publisher contracts have started including AI-specific language, and not all of it favors authors. Before signing, you need to know which clauses protect your rights, which ones quietly waive them, and where publishers are inserting leverage that did not exist two years ago.

The Authors Guild Model Clauses

The Authors Guild has published model contract language designed to give authors a baseline for negotiation. The core AI-training clause explicitly reserves all AI training rights to the author: "For avoidance of doubt, except as expressly provided for in this Agreement, Author reserves the rights, and Publisher acquires no rights to, reproduce, distribute and/or otherwise use or create derivative works of the Work, or to sublicense any of the foregoing, for purposes of training generative artificial intelligence technologies (AI) or for use in any AI outputs; provided that Publisher may use the Work without the Author's consent or approval in connection with internal AI-powered tools and technologies utilized in the normal course of Publisher's internal operations (including but not limited to improving the marketing, advertising, and searchability of the Work)." That carveout for internal publisher operations matters — it is narrowly drawn, but it is still a concession. Know what you are agreeing to.

On author use of AI, the Guild's model language draws a hard cap: "Author shall not be required to use generative AI or to work from AI-generated text. Author shall disclose to Publisher if any AI-generated text is included in the submitted manuscript, and may not include more than [a de minimis/5%] AI-generated text." This is the standard most traditional publishers will likely migrate toward. The bracket around "de minimis/5%" signals that this is a negotiation point — push for the language that applies to your process. And the Guild's protections run in both directions: "Publisher agrees and warrants that it will not use AI to substantially edit a manuscript (excepting the use of basic spelling and grammar-checking applications)."

If a publisher insists on licensing AI training rights rather than having the author fully reserve them, the Guild recommends specific revenue splits as a floor: 80–90% to the author for AI Training Only and for Character Chat or Fan Fiction uses; 50/50 for RAG, AI Summary, or Chat Features. These numbers should anchor your negotiation if a publisher refuses a full reservation. You can review the Authors Guild's full model AI clauses to use them verbatim or adapt them for your contract redline.

The New Termination Risk: AI Warranty Clauses

Watch for a newer clause type that has appeared in some publisher contracts: an AI warranty that allows cancellation if the publisher believes the author breached an originality or "no AI" warranty. The danger is in that word — believes. Some contracts permit publishers to act on AI detector scores alone, despite the fact that these tools produce unreliable results and have flagged human-written prose as AI-generated. If a publisher insists on an AI warranty, negotiate to require documented, material proof of undisclosed AI use — not mere suspicion and not a detector score. Suspicion-based termination clauses shift all the risk to authors while giving publishers an off-ramp from inconvenient deals.

Practical Steps Before You Submit or Sign

Documentation is your first line of defense. Keep detailed records of your creative process — timestamps, draft versions, notes on structural choices, revision history. If the Copyright Office or a publisher ever questions the human-authorship basis for your work, contemporaneous records are what make that case. A folder of versioned drafts and a short process log cost nothing to maintain and can matter enormously later.

Disclosure is not optional, and the sooner you make it, the less risk you carry. If AI-generated text appears anywhere in a manuscript you are submitting for copyright registration, inform your publisher so the registration application is accurate — filing a materially incomplete application creates problems that are far harder to fix after the fact. For reader-facing disclosure, the Authors Guild recommends describing AI use in the front matter of a book or at the bottom of an article — a sentence or two, placed where readers will see it.

The same transparency principle applies to anyone you hire. If you work with a freelance editor, developmental editor, or writing contractor, your written contract should prohibit the addition of more than de minimis AI-generated text without your written approval and require that they disclose their use of AI. Without that clause, a contractor could introduce AI content into your manuscript without your knowledge — and the copyright and contractual consequences land on you, not them.

Before signing with an agent or submitting to a publisher, have a direct conversation. Let your agent and publisher know up front that you have incorporated AI outputs, and ask whether the publisher has specific guidance. That conversation belongs at the beginning of the relationship, not surfaced during contract negotiation or after the manuscript is accepted.

Publishing contracts are where the abstract becomes binding. AI warranty clauses, indemnification language, and rights provisions are negotiable — but only before you sign. If you are entering a publishing agreement for a work that involved any AI-assisted writing, have the contract reviewed by counsel who understands intellectual property and the current publishing landscape before you execute it.