AI and copyright sit in an awkward place under U.S. law: work produced entirely by an AI system is not protected and enters the public domain the moment it’s generated, while work that combines human creativity with AI-generated material can be protected in part. Only the human-authored elements qualify. How much of your work is actually covered depends on how much creative control you exercised over the final result, and on whether you disclose the AI-generated portions when you register.
Why Pure AI Output Gets No Protection
Federal copyright law protects works of human authorship. The Supreme Court has defined an author as “he to whom anything owes its origin; originator; maker; one who completes a work of science or literature.”1Constitution Annotated. ArtI.S8.C8.3.1 Authorship, Writings, and Originality The Copyright Office will not register works produced by a machine or mechanical process operating without creative input from a human author.2Federal Register. Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence
That position has now been tested and upheld in court. Stephen Thaler tried to register an image created entirely by his “Creativity Machine” AI system, naming the machine as the author. The Copyright Office refused. In March 2025, the D.C. Circuit affirmed the refusal, holding that “the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being.”3United States Court of Appeals for the District of Columbia Circuit. Thaler v. Perlmutter The Supreme Court declined to hear the case in March 2026, leaving that ruling in place.4Supreme Court of the United States. Docket for 25-449
The practical result: if you generate an image or a block of text from an AI tool and do nothing else with it, you have no copyright in the output. Anyone can copy it, modify it, or sell it. You cannot send a DMCA takedown for something you don’t hold a copyright in, and a competitor using the same AI-generated content is not infringing anything.
When AI-Assisted Work Is Protected
The analysis changes when a human contributes real creative expression. The Copyright Office’s decision on the graphic novel Zarya of the Dawn is the clearest example. Author Kristina Kashtanova used Midjourney to generate the book’s illustrations. The Office found she owned a valid copyright in her written text and in her creative selection, coordination, and arrangement of the book’s written and visual elements. The individual AI-generated images, though, were not copyrightable because they were “not the product of human authorship.”5United States Copyright Office. Zarya of the Dawn (Registration # VAu001480196)
Treat AI-generated content the way you would treat raw material from the public domain. You can build something original around it, and your original contribution is protected, but the underlying AI material stays free for anyone else to use. Substantial modification of the output, or its incorporation into a larger human-authored work, can also support protection. The Copyright Office has recognized that “creative arrangements or modifications of the output” can qualify.6U.S. Copyright Office. Copyright Office Releases Part 2 of Artificial Intelligence Report
Why Writing a Good Prompt Isn’t Enough
Many creators assume that a carefully engineered prompt gives them authorship over what the AI produces. The Copyright Office disagrees. In its January 2025 report, the Office concluded that “prompts do not alone provide sufficient control” to make users the authors of AI output, because prompts “essentially function as instructions that convey unprotectible ideas” rather than dictating a specific creative result.7U.S. Copyright Office. Copyright and Artificial Intelligence Part 2: Copyrightability
Iterative prompting doesn’t change the outcome. Refining and resubmitting prompts until you get a result you like is, in the Office’s view, closer to re-rolling the dice than to authoring the image. The final output still reflects the system’s choices, not the user’s.7U.S. Copyright Office. Copyright and Artificial Intelligence Part 2: Copyrightability The Zarya decision drew the analogy that a person prompting Midjourney is more like a client giving general directions to a commissioned artist than the artist doing the actual painting.5United States Copyright Office. Zarya of the Dawn (Registration # VAu001480196) Time spent perfecting a prompt doesn’t help, because copyright protects expression, not effort.
How to Register a Work That Includes AI Content
If your work combines human-authored and AI-generated material, register it through the Copyright Office’s Electronic Copyright Office (eCO) system. The extra work is in separating your contribution from the AI’s.
On the Standard Application, use the “Author Created” field to describe what you personally created, such as “text written by author” or “selection and arrangement of visual elements.” In the “Limitation of Claim” section, use the “Material Excluded” field to disclaim the AI-generated portions and briefly identify what the AI produced, for example, “images generated by [name of AI tool].” AI-generated content that is more than de minimis must be explicitly excluded from the claim.2Federal Register. Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence
Filing fees range from $45 for a single-author, single-work claim to $65 for a Standard Application.8U.S. Copyright Office. Fees Payment comes before the system prompts you to upload copies of the work.9U.S. Copyright Office. Online Registration Help (eCO FAQs) Online applications with digital uploads that don’t require examiner correspondence average about two months to process. Claims that require examiner correspondence, which is a real possibility once AI use is disclosed, average around four months and can run past eight.
What Happens If You Don’t Disclose the AI
Disclosure is not optional. If you’ve already registered a work without disclosing AI use, file a supplementary registration to correct the record, describing your human contributions in the “Author Created” field and disclaiming the AI-generated material in “Material Excluded.”10U.S. Copyright Office. Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence
Leaving the record uncorrected carries real consequences. If the Copyright Office learns that essential information about AI use was omitted or is questionable, it can cancel the registration.10U.S. Copyright Office. Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence In an infringement suit, the opposing party can also challenge the registration under the Copyright Act. A court will disregard the registration if the applicant knowingly gave inaccurate information and the accurate information would have led the Office to refuse registration.11Office of the Law Revision Counsel. United States Code Title 17 Section 411 – Registration and Civil Infringement Actions
Knowingly making a false statement of material fact on a copyright application is a federal offense punishable by a fine of up to $2,500.12Office of the Law Revision Counsel. United States Code Title 17 Section 506 – Criminal Offenses The larger cost is enforcement. If your registration is invalidated, you lose access to statutory damages and attorney’s fees in infringement litigation, which for a commercial creator is a much heavier blow than the fine.
What Platform Terms Like “You Own the Output” Actually Mean
Most AI platforms assign users whatever ownership rights the company may have in the generated output. OpenAI’s terms of use state that “you own the Output” and that the company assigns “all our right, title, and interest, if any, in and to Output.”13OpenAI. Terms of Use Read the qualifier. OpenAI is transferring whatever rights it might hold, and under current copyright law that is nothing. You can’t assign rights that don’t exist.
The terms also don’t promise exclusivity. OpenAI acknowledges that other users may receive similar or identical output from the same service, and the assignment doesn’t reach what other users generate. Someone else could produce a near-identical image, and neither of you would hold a copyright in it. Contractual terms with the platform matter for disputes between you and the company; they do not create copyright protection where federal law says there is none. If you’re using AI output commercially, your protection against copying comes from contract, trade secret, or trademark law, not from copyright.
A Separate Question: AI Training on Copyrighted Works
Whether AI companies can legally train their models on copyrighted books, images, and other material is a different copyright question and is not settled. Several major lawsuits are working through the federal courts. Early rulings from 2025 have gone in different directions depending on the facts: some courts have found training to qualify as fair use when the purpose is highly transformative and the plaintiff cannot show market harm, while others have distinguished between training on lawfully purchased material and training on pirated copies. No appellate court has issued a definitive ruling, and the outcomes have been heavily fact-specific. For creators concerned about their existing work being used as training data, the law here is genuinely open.