The Copyright Office ruled prompts alone do not make an author

Part 2 of the Office's AI report finds prompting insufficient for authorship, with consequences for songs built on generated parts.

Historical event

January 29, 2025

First source published

January 29, 2025

Site publication

September 18, 2026

What happened

On 29 January 2025 the United States Copyright Office published Part 2 of its Copyright and Artificial Intelligence report, the middle instalment of a three-part study launched after a 2023 notice of inquiry. Part 1, on digital replicas of voice and likeness, appeared in 2024; a third part on training data followed later in 2025. Part 2 answers one question: when does a work made with a generative system qualify for copyright registration. The Office's own announcement frames the report as applying existing law rather than recommending new legislation, a position the report repeats in its conclusions.

What the documents say

The report states plainly that "the case has not been made for additional copyright or sui generis protection for AI-generated content," and that "based on the functioning of current generally available technology, prompts do not alone provide sufficient control" to support a claim of human authorship. It draws a line rather than a ban: copyright still protects the human-authored expression in a work even where AI-generated material is also present, and it protects a human's creative selection, arrangement or modification of generated output. What it withholds protection from is material where a system, not a person, made the expressive choices. The report records comments from the Recording Academy that members already use generative tools for song ideation and from Universal Music Group on production workflows, treating that kind of use as assistive rather than as authorship in itself. The Office's landing page, current as retrieved on 16 September 2026, confirms the report is one of three parts and that the Office intends to keep monitoring practice.

Why it matters for makers

A producer who types a prompt and accepts the output has, on the Office's reasoning, made nothing registrable by that act alone. What earns protection is what a human adds afterwards: an arrangement, an edit, a chosen combination of stems, a rewritten lyric. That distinction matters commercially, because an unregistrable element cannot be licensed, sold or enforced against a copyist the way a composition or master can. A track built mostly from unedited generation carries a thinner rights package than one where a session's human choices are documented.

What to check before you use it

Keep a record of what a human changed after generation: edits, arrangement, mixed-in performances. This is editorial guidance beyond the report itself, but it follows from the distinction the Office draws between assistance and authorship. Before registering a track with generated parts, identify which elements are claimed as human authorship and be ready to describe the process.

• Which parts of this track were shaped by a human decision after the system generated its output?
• Could I describe, element by element, what I contributed if a registration were challenged?
• Does my collaboration agreement say who owns the human-authored edits layered onto a generated draft?

The report does not ban generative tools from a session; it withholds registration from the parts of a work where no human made the expressive choices, which is a narrower and more useful test than either "AI is unprotectable" or "AI is just another instrument."

Sources & reading trail

States the Office's conclusions, including that prompts alone do not provide sufficient control for authorship and that human-authored elements remain protectable.

Source published: 29 January 2025 · Retrieved: 16 September 2026

Summarises the report's release and quotes the Register on when AI-assisted creativity is protectable.

Source published: 29 January 2025 · Retrieved: 16 September 2026

Confirms the three-part report structure and publication dates for Parts 1-3, current as retrieved.

Source published: Not established · Retrieved: 16 September 2026

Papers, terms and official documents establish the record; the maker reading and the checks are Signal to Song editorial analysis. This retrospective draft does not imply the site published on the event date.

Continue reading

• Courts ruled a fully AI-made work has no human author
• The Copyright Office found prompts do not control output
• Open models show two ways to disclose training data
• Browse the complete the archive

Published September 18, 2026, not on the date of the event described.