Copyrightability

AI and Copyright Law: The Complete 2026 Overview

By Nora Whitfield · August 23, 2026

AI and Copyright Law: The Complete 2026 Overview

If you're trying to figure out where AI and copyright law actually stand right now, you're wading into one of the fastest-moving corners of intellectual property law in a generation. Courts have been ruling on it monthly. The Copyright Office has published a three-part report on it. And the answer to "can I copyright something I made with AI" is neither a clean yes nor a clean no. It depends on what you did.

Here's the state of things in plain English, as of 2026.

The one rule everyone agrees on: a human has to be the author

Start with the floor, because everything else builds on it. In Thaler v. Perlmutter, a computer scientist named Stephen Thaler tried to register a copyright listing his AI system as the sole author, with himself owning the work only because he owned the machine. The Copyright Office said no. Thaler sued, lost at the district court, lost again at the D.C. Circuit in March 2025, and on March 2, 2026, the Supreme Court denied his petition for review. That ends it. There is no path to copyrighting a work with zero human authorship, and there won't be one without an act of Congress.

This is the part people get right. The part that trips people up is what comes next: assuming that if AI touched your work at all, you're automatically in Thaler territory. You're not. Thaler removed the human on purpose, as a test case. Almost nobody else does that.

The real question: how much did a human actually shape it

Once you clear the "is there a human at all" bar, you're in the territory that actually governs most AI-assisted creative work: how much of the expression did a person control?

The Copyright Office has been building this standard case by case through actual registration decisions, and a few have become the reference points everyone cites. In Zarya of the Dawn, a graphic novel with Midjourney-generated art, the Office protected the text and the way the author arranged the images into a story, but refused to protect the individual images themselves, since Midjourney made the creative choices about what those images looked like. In Théâtre D'opéra Spatial, a Colorado State Fair winner wrote 624 prompts to refine an image and the Office still said no, ruling that prompting, however extensive, is instruction to a tool rather than direct control over the expression. Compare that to A Single Piece of American Cheese, where a human selected specific AI outputs and arranged them into a finished composition in a way the Office found registrable.

The pattern across all three: prompting alone doesn't cut it, no matter how detailed. What counts is a human determining the actual expressive elements, whether that's writing the words yourself, meaningfully editing AI output, or making creative arrangement and selection decisions that go beyond picking a favorite from a list.

The Copyright Office laid this out formally in its January 2025 report on copyrightability, the second part of a three-report series examining AI's collision with copyright law from different angles: digital replicas, copyrightability of outputs, and how training data should be handled. If you want the full breakdown of that report, we've written a walkthrough of all three parts.

The other fight: is training AI on copyrighted books and art legal at all

Separate from whether your AI-assisted output is copyrightable, there's a second, much bigger legal war happening over whether AI companies were allowed to train their models on copyrighted material in the first place. This is where the highest-stakes money is.

The New York Times sued OpenAI and Microsoft over training data and alleged output reproduction. A group of authors, including household names, brought a consolidated case against OpenAI. Getty Images sued Stability AI in both the UK and US over image training data. Anthropic settled with a group of authors for $1.5 billion after a court found some of its training practices infringing while other uses were protected as fair use, a split ruling that neither side got to claim as a clean win. Music publishers went after Suno and Udio directly. This list keeps growing, and we track the full docket in our complete guide to every AI copyright lawsuit worth knowing.

Why does this matter to you if you're not an AI company? Because the fair use question and the authorship question are getting tangled together in public conversation, and they're not the same fight. Whether OpenAI infringed by training on the New York Times' archive has nothing to do with whether the essay you wrote with ChatGPT's help is yours to copyright. Keep those two questions separate in your head, because a lot of confused commentary doesn't.

What a normal person using AI in creative work actually needs to know

If you're a writer, artist, developer, or musician who uses AI tools as part of your process, here's what actually matters, cutting through the noise:

Your prompts alone won't get you a copyright, no matter how clever or detailed they are. The expressive decisions need to be yours: what you write yourself, how you edit and revise AI-generated material, how you select and arrange pieces into something new, and any creative choices layered on top of a raw AI output. A blog post you wrote and had AI clean up the grammar on is yours. A novel you generated wholesale from a plot outline prompt, with no rewriting, is a much harder case.

Keep records as you go. Not because you're required to submit them with your application, but because if your authorship is ever challenged, "I did the creative work" is a much stronger claim backed by a contemporaneous record than a reconstruction after the fact. That's a subject worth its own deep dive, which you'll find in our guide to how to prove human authorship of AI-assisted work.

Disclose AI use honestly on your application. The Copyright Office wants you to identify AI-generated material you're excluding from your claim, and getting this wrong, whether by omission or overclaiming, can jeopardize the registration. We cover the specifics in our guide to what the Copyright Office requires for AI-assisted work.

Where this is headed

Nothing here is fully settled. The fair use cases are still working through appeals. Congress has floated multiple bills aimed at forcing AI companies to disclose training data, though none have become law yet. The Copyright Office itself has had leadership turnover during this period that's added its own layer of uncertainty about how consistently the guidance will be applied. If you're building something with AI as part of your process, the safest posture is the one that's worked since well before any of this litigation started: do real creative work, keep track of what you did, and don't assume a prompt is a substitute for authorship.