AI and copyright: what you can own, use and must disclose

AI tools now sit inside the same apps people use to write, draw, compose and edit. The copyright questions arrive just as fast: can I own this, who else might, do I have to say I used AI, and how do I show the work is mine? This section answers those questions in plain English, using the documents the US Copyright Office and the federal courts have actually published.

Each page names its sources and their dates, and each one says where the law is still unsettled. If you want the short version, read the sections below. If you want the detail, the full list of pages is at the bottom.

01The rule in one paragraph

US copyright covers "original works of authorship," and the Copyright Office reads "author" to mean a human being. In its March 16, 2023 registration guidance in the Federal Register, the Office said material an AI system generates on its own is not protected and should be excluded from a registration claim, while the parts a person creates, selects, arranges or meaningfully changes stay protected. On January 29, 2025, the Office's report on copyrightability added that prompts alone do not give a person enough control over the result, based on the tools generally available today.

For the full test, with the cases that shaped it, read our guide to whether AI-generated content can be copyrighted.

02Owning what you make with AI

There are two kinds of ownership to keep apart. The first is a contract: the terms of service you accepted when you signed up for the tool. Many tools say they hand over whatever rights they have in the output. The second is copyright, which comes from the law, not from the tool. A company can only give you rights it actually has.

So you can be the only party with a contract claim to an image and still hold no copyright in it. Our page on who owns AI output under the tools' own terms quotes those terms, dated, and explains what they can and cannot do for you.

03What the Copyright Office has said

The Copyright Office has published more on AI than any other US body. Its key documents, in order:

• March 16, 2023: registration guidance for works containing AI-generated material.
• August 30, 2023: a notice of inquiry, which drew more than 10,000 public comments.
• July 31, 2024: Part 1 of its report, on digital replicas (deepfakes of real people).
• January 29, 2025: Part 2, on copyrightability of AI outputs.
• May 9, 2025: Part 3, on AI training, released as a pre-publication version.

We summarize each one in the Copyright Office's AI guidance, in plain English. The decision that started it all was a comic book. Our page on the Zarya of the Dawn decision explains what the Office kept and what it cut.

04Training and fair use

Whether you can own AI output is one question. Whether an AI company may copy your work to train its system is a different one. Training usually involves making copies, so the main legal question is fair use: the four-factor test in Section 107 of the Copyright Act.

The Copyright Office's Part 3 report, released in pre-publication form on May 9, 2025, applies those four factors to training. Courts are deciding the question case by case. Our guide to AI training, fair use and opting out covers what creators can do now.

05Disclosure

"Disclose AI" can mean three different things, and people mix them up:

1. To the Copyright Office. If you register a work that contains more than a trivial amount of AI-generated material, the March 16, 2023 guidance says to describe it and exclude it from your claim. See how to disclose AI on a copyright application.
2. To a platform. Publishing platforms such as Amazon KDP ask their own questions about AI content, with their own definitions.
3. To readers, teachers or clients. This is about honesty and sometimes about rules where you live. Our page on writing an AI disclosure statement has examples by medium.

Separate from all three is the school and workplace question. Using AI can break an honesty rule without breaking any law. We explain the difference in whether using AI counts as plagiarism.

06Proving you made it

The reverse problem is growing too: people who wrote or drew something themselves are told it looks machine-made. Copyright does not require you to prove authorship in advance. But when a teacher, publisher or platform asks, records help.

Drafts, version history, sketches, notes and source files are the usual starting point. None of these is proof on its own, but together they tell a story a reviewer can follow. Read more in how to show evidence of human authorship. Newer tools can also attach a signed record of how a file was made; see content credentials and C2PA labels.

07Where the law is still moving

Some parts of this area are settled, and some are not.

• Settled for now: a machine cannot be the author. The D.C. Circuit held this in Thaler v. Perlmutter on March 18, 2025, and the Supreme Court denied review on March 2, 2026 (docket No. 25-449). For the case history, see the Thaler v. Perlmutter case file.
• Still open: how much human control is enough. Jason Allen's challenge to the refusal of his Midjourney image, Allen v. Perlmutter, was filed in federal court in Colorado on September 26, 2024. We found no ruling on the merits as of October 30, 2026.
• Still coming: a final version of the Part 3 training report, and updated registration guidance. On January 29, 2025, the Office said it plans to supplement its March 2023 guidance and update its Compendium of practices.

No prompt count, edit percentage or tool setting guarantees registration; the Copyright Office decides human authorship case by case.