What the US Copyright Office says about AI, in plain English

What has the US Copyright Office actually said about AI? More than most people realize, spread across a policy statement, a three-part report and a series of registration decisions. This page puts it in one place, in date order, in plain English.

It belongs to our section on AI and copyright law. Every date here comes from the Copyright Office's Copyright and Artificial Intelligence page. At the end is a decision table you can use to see where your own work probably falls.

01The March 2023 registration guidance

On March 16, 2023, the Office published a statement of policy in the Federal Register: Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence (88 FR 16190). It took effect the same day and is still the Office's main rule for applicants.

The test

The Office asks whether a work is "basically one of human authorship, with the computer merely being an assisting instrument," or whether its traditional elements of authorship were produced by a machine. It called this "necessarily a case-by-case inquiry."

The duty to disclose

The guidance says applicants "have a duty to disclose the inclusion of AI-generated content" in a work they submit, with a brief explanation of what the human author contributed. AI material that is more than de minimis (more than trivial) should be excluded from the claim. Applicants should not list an AI tool, or the company behind it, as an author or co-author.

Fixing older filings

The guidance also explains how to correct a pending application or an issued registration that left AI material out. It warns that applicants who do not correct the record "risk losing the benefits of the registration." Step-by-step instructions for each field are on our page about how to complete the AI disclosure on a copyright application.

02Part 1: digital replicas

The Office's larger study began with a notice of inquiry on August 30, 2023, which drew more than 10,000 comments. The resulting report is being issued in parts.

Part 1, published July 31, 2024, covers digital replicas: video, images or audio made or altered to "realistically but falsely depict an individual." Deepfakes, in other words. The Office found gaps in existing law and recommended that Congress pass a new federal law protecting all individuals against the knowing distribution of unauthorized digital replicas.

Part 1 is about people's likenesses and voices, not about who owns AI output. It matters if someone makes a fake of you. It does not change the registration rules.

03Part 2: copyrightability

Part 2, published January 29, 2025, is the one most creators need. It addresses whether AI outputs can be copyrighted. Its main conclusions, as the Office summarized them:

• Questions of AI and copyrightability can be resolved under existing law, without new legislation.
• Using AI to assist, rather than stand in for, human creativity does not affect protection.
• Copyright protects a human's original expression even if the work also includes AI material.
• Copyright does not extend to purely AI-generated material, or material where a person had too little control over the expressive elements.
• Based on current generally available technology, prompts alone do not provide enough control.
• Human authors are entitled to copyright in their own work that is perceptible in an AI output, and in creative selection, arrangement or modification of output.
• The case has not been made for a new kind of protection for AI-generated content.

The Register of Copyrights, Shira Perlmutter, put it this way in the Office's January 29, 2025 announcement: "Extending protection to material whose expressive elements are determined by a machine, however, would undermine rather than further the constitutional goals of copyright." For how this plays out in practice, read can you copyright AI-generated work.

04Part 3: training, still pre-publication

Part 3 deals with a different question: whether AI companies may use copyrighted works to train their systems. The Office released it on May 9, 2025 as a "pre-publication version," in response to congressional inquiries and public interest. As of October 2026, the Office's AI page still lists it as pre-publication. It says a final version "will be published in the future, without any substantive changes expected in the analysis or conclusions."

Part 3 describes how training works, explains why it can involve copying, and treats fair use as the main defense. It applies the four factors in Section 107 of the Copyright Act. One question it weighs is whether AI output can dilute the market for the works a system learned from. It also looks at voluntary licensing as a way forward. When the Office released it, the report noted that dozens of lawsuits on the question were pending in the United States. Courts, not the Office, will decide them.

Part 3 does not change the registration rules for your own work. It matters if you want to know whether your work can be used to train AI without permission.

05What the courts have added

The Office's guidance is not the last word; courts can review its decisions. So far they have agreed with it on the core point.

• Thaler v. Perlmutter. On March 18, 2025, the D.C. Circuit held that the Copyright Act "requires all eligible work to be authored in the first instance by a human being." The Supreme Court denied review on March 2, 2026. The court did not decide how much human input is enough, because Thaler had named his AI as the sole author.
• Allen v. Perlmutter. Artist Jason Allen sued in federal court in Colorado on September 26, 2024, after the Office's Review Board refused his Midjourney image on September 5, 2023. The case asks whether detailed prompting can be authorship. We found no ruling on the merits as of October 30, 2026.

The registration decision that first applied these ideas to a mixed work was a comic book; see the Zarya of the Dawn ruling explained.

06A decision table for your work

This table applies the Office's March 16, 2023 guidance and its January 29, 2025 report to common situations. It is a starting point, not a prediction: the Office decides each application on its facts.

If someone later questions whether you made the work at all, the records you kept matter more than any form. See what counts as evidence of human authorship.