//Owning what AI helped make
Who Owns Work That AI Helped Make
If artificial intelligence helped make it, a United States professional owns the part a human being actually created. That is where the United States Copyright Office and the District of Columbia Circuit have both landed, and the practical rules that follow from it are clearer than most people think. Every source below is named and dated.
By Matt Aubin, Founder, Southern Recon Agency and E3 Legacy Intel. CDFE, FBCI. Investigating since 2009. Updated .
Read this first
This page is teaching, not legal advice. It summarizes public federal sources, names each one and dates it, and it does not create a lawyer client relationship. Ask a lawyer about any specific matter, including any decision about registration, ownership or AI use at your own firm.
Every source named on this page is a federal one: the Copyright Office, the United States Code, decided court opinions, the federal standards body and the Federal Trade Commission. Each claim below carries the source it came from, with its date.
Can you copyright AI assisted work?
Yes, to the extent a human made the creative choices in it. The United States Copyright Office said in its registration guidance at 88 Fed. Reg. 16190, published March 16, 2023, that based on the functioning of current generally available technology, when an AI technology determines the expressive elements of its output, the generated material is not the product of human authorship. Its report Copyright and Artificial Intelligence, Part 2: Copyrightability, published January 29, 2025, states the other half: existing law already answers these questions, AI used as an assisting tool does not cost an author protection, and prompts do not alone provide sufficient control. The human authored elements of an AI assisted work remain protectable; the machine determined elements do not.
Who owns AI generated content, code and images?
Under 17 U.S.C. 201(a) copyright vests initially in the author of the work, and under 201(b) an employer is treated as the author of a work made for hire unless a signed agreement says otherwise. The catch is the word author. The District of Columbia Circuit held in Thaler v. Perlmutter, No. 23-5233, decided March 18, 2025, that the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being, while making clear that the rule does not prohibit copyrighting work that was made by or with the assistance of artificial intelligence. Rehearing was denied on May 12, 2025, and the Supreme Court denied the petition for review in No. 25-449 on March 2, 2026, which left the Circuit's decision standing without the Supreme Court deciding anything itself. A denial of review is not a ruling on the merits. What follows for a working professional is that purely machine generated content, code or images have no human author to own them, while the human written, human chosen and human edited parts around them do.
The cleanest worked example is the Copyright Office decision on the comic Zarya of the Dawn, Reg. VAu001480196, dated February 21, 2023. The author kept copyright in her written text and in her creative selection and arrangement of the material. The images inside the book were held not to be the product of human authorship, even after hundreds of prompts.
What do you have to tell the Copyright Office?
Less than most people assume, and it depends on how much AI generated material the work contains. Section IV of the same March 16, 2023 guidance at 88 Fed. Reg. 16190 asks an applicant to disclose the inclusion of AI generated content that is more than a trivial amount, to identify the human author's own contribution, and to exclude the AI generated material from the claim. The guidance is equally clear that in some cases nothing needs to be disclaimed, because material that is trivial in amount does not have to be carved out. So the question a registrant faces is how much of the expression the machine determined, not whether AI was in the room at all.
What happens if you already registered something and did not disclose it?
The same March 16, 2023 guidance directs applicants to correct the public record rather than leave an incomplete registration standing, and it sets out the route for doing so. It also states that a registration can be cancelled where the applicant does not respond to the Office on the point. The practical shape of it is that the correction is an administrative step while the omission goes to the registration itself. What to do about any particular filing is a question for a lawyer, not for this page.
Is it legal to train AI on other people's work?
Genuinely unsettled, and the split results are the proof. In Bartz v. Anthropic, No. 3:24-cv-05417 in the Northern District of California, the court held on June 23, 2025 that training on lawfully obtained books was exceedingly transformative and was a fair use, while keeping a permanent library of pirated books was not excused; the court granted final approval on July 20, 2026 to a settlement fund of 1.5 billion dollars over the pirated copies, an objector appealed on August 31, 2026, and the settlement has not been paid out. In Kadrey v. Meta, No. 23-cv-03417-VC, decided June 25, 2025, Meta won on fair use only because those plaintiffs presented no meaningful evidence on market dilution at all, and the judge warned that in cases involving uses like Meta's, it seems like the plaintiffs will often win, at least where a case has a better developed record on the market effects of the use. In Thomson Reuters v. Ross, No. 1:20-cv-613-SB in Delaware, decided February 11, 2025, copying legal research summaries to build a competing research tool was held not to be fair use; the appeal was argued June 11, 2026 and is undecided. Andersen v. Stability AI and New York Times v. Microsoft and OpenAI are both live, and neither has a fair use ruling on its docket.
The Copyright Office itself has said, in Part 3 of its report on generative AI training, that using vast troves of copyrighted work to produce content that competes with it, especially where this is accomplished through illegal access, goes beyond established fair use boundaries. Part 3 is still the pre publication version. There is no 2026 report from the Office.
Get it written down before somebody asks.
Where AI is used in your work, who checked it, and what was disclosed, in one signed document renewed every year.
What about trade secrets and AI?
This is the exposure that does not need a court to bite. Federal law at 18 U.S.C. 1839(3) protects information as a trade secret only where the owner thereof has taken reasonable measures to keep such information secret, which makes how material is handled part of whether it is protected at all. The federal standards body treats the risk as a named one: its generative AI profile, NIST AI 600-1, published July 2024, lists intellectual property among the risks specific to generative AI and expressly covers eased exposure of trade secrets. Whether any particular handling practice meets the statute is a question for a lawyer, and this page does not answer it.
Can you get in trouble for how you describe your AI?
Yes, and the enforcement is already live. The Federal Trade Commission announced Operation AI Comply on September 25, 2024, bringing five cases against companies that used AI claims or AI tools to deceive people, and said there is no AI exemption from the laws on the books. One of those five orders was set aside in December 2025. The Commission has kept going: orders finalized in August 2026 required three firms to pay 930,000 dollars over an AI marketing service. The practical lesson for a professional firm is that what you say your AI does is an advertising claim, and it is treated as one.
What does this mean for a firm in practice?
It turns ownership into a record keeping question as much as a legal one. The human contribution is the part that is protectable, so a firm whose files show what a person actually wrote, chose and edited is in a different position from one whose files show only a finished document. Handling matters for the same reason, because the trade secret statute turns on the measures an owner took. And the sources above are all ones an insurer, a client or opposing counsel can read, which is why firms tend to want one dated document stating where AI is used in their work, who checks its output, and what was disclosed, rather than assembling that answer under pressure.
That document is the signed AI evidence audit, measured against the Evidence-Grade AI Standard. Teaching your people the underlying skill is prompting training.
//Why a practitioner wrote this
Written by somebody who has to live with the answer.
Investigating since 2009. Founded his first firm in 2010 and built it into Southern Recon Agency in 2014. He built a proprietary AI investigation platform his own firms run on live cases, so the ownership and handling questions on this page are ones he had to settle inside his own business before he taught anyone else. That is the only claim being made here: the sources are public, and the reading of them is a working practitioner's.
He teaches the tricks of the trade. He does not give legal advice, and nothing here is a substitute for asking a lawyer about your own matter.
The line, again, at the bottom of the page
This page is teaching, not legal advice. It summarizes public federal sources, names each one and dates it, and it does not create a lawyer client relationship. Ask a lawyer about any specific matter, including any decision about registration, ownership or AI use at your own firm.
If the question is whether a piece of AI assisted evidence will hold up, that is AI evidence consulting. The framework behind the audit is Evidence-Grade AI, and the way to teach your own people is prompting training.
Put the answer in writing before the question arrives.
Tell me how AI touches your work today. You get a straight answer on whether it is the signed audit, a training day, or nothing at all yet.