You prompt Claude to write a 2,000-word article. You prompt Midjourney to create a painting. You use ChatGPT to help write your novel. You generate code with GitHub Copilot and ship it to paying customers. In every case, a reasonable question arises: who owns this? The answer in 2026 is more complex, more nuanced, and in some ways more practical than the breathless coverage would suggest. This guide walks through every major dimension of AI copyright — what you own, what you do not, how to protect your position, and the legal battles still actively shaping this landscape.
The Core Principle: US Copyright Requires Human Authorship
The foundational legal principle governing AI-generated content in the United States is clear and has been reaffirmed multiple times: US copyright law requires human authorship. Works generated entirely by AI — without meaningful human creative input — are not eligible for copyright protection. They are, by default, in the public domain. This has been the consistent position of the US Copyright Office in multiple rulings since 2023, including the 2025 AI copyright guidance that clarified the 'human authorship' standard.
The Spectrum: From 'No Copyright' to 'Fully Copyrightable'
AI content ownership is not a binary — it exists on a spectrum based on the nature and degree of human creative input. Understanding where on this spectrum your work falls determines your legal position.
- Pure AI output (no human authorship): you type 'write a poem about autumn' into ChatGPT, it produces a poem, and you publish it with no modification. The Copyright Office's position is clear: this is not eligible for copyright protection. Anyone can legally copy, modify, and use it without consequence. The poem is in the public domain the moment it is generated.
- Lightly prompted AI output with minor editing: you write a detailed prompt specifying style, structure, tone, and content, generate the output, make minor edits, and publish. The Copyright Office has been inconsistent in this middle zone. The guidance suggests copyright is possible if there is meaningful human creative selection and arrangement — but 'meaningful' is not precisely defined.
- AI-assisted human creative work: you write a first draft of an article, use AI to suggest improvements and fix grammar, then revise substantially based on those suggestions. Your own original expression — the parts you wrote and the substantive creative decisions you made — are copyrightable. The AI-generated suggestions that you incorporated, as transformed by your editorial judgment, may also be protected as part of the whole work.
- AI-generated elements selected and arranged by humans: you generate 100 AI images and curate 10 of them into an art book with your commentary and arrangement. The AI images themselves may not be individually copyrightable, but the selection, arrangement, and compilation — as original creative choices — may qualify for 'compilation copyright,' which protects the organization and selection even when individual elements are not protected.
- Prompt engineering as protected expression: in some cases, the prompt itself may qualify for copyright protection as original creative expression. If your prompt is a detailed, creative, original piece of writing, it may be protected — even if the output is not.
The Landmark Court Decisions That Define the Landscape
- Thaler v. Vidal (2022–2023): a key early case establishing that AI systems cannot be named as inventors on patents. Applied to copyright: only humans (or corporations as legal persons) can hold copyright. The AI system that generates a work cannot be the copyright holder.
- The Sarah Andersen et al. v. Stability AI case: class action lawsuit by visual artists arguing their copyrighted works were used without permission to train Stability AI's image generation models. This litigation is ongoing and addresses the training data question rather than the output ownership question — but its resolution will significantly affect AI company liability for training data.
- The New York Times v. OpenAI: the NYT alleges that its articles were used to train ChatGPT without permission and that ChatGPT reproduces NYT content too closely in some responses. Ongoing. The outcome could establish significant precedent for how training data use is treated under US copyright law.
- Copyright Office 2025 AI guidance: the most recent comprehensive guidance from the Copyright Office establishes that 'where the traditional elements of authorship are conceived and executed by machine, the resulting work lacks human authorship and the Office will not register it.' This guidance does not resolve all edge cases but provides a clear baseline.
Practical Questions Answered: What You Can and Cannot Do
- Can I sell AI-generated art? Yes, in most cases — but you may not be able to claim copyright, which means you cannot legally prevent others from copying and selling the same image. If you generated it with Midjourney, Midjourney's Terms of Service (for paid subscribers) say you own the commercial rights to use the output, but this is a contractual right from Midjourney, not a copyright.
- Can I publish AI-written content under my name? Yes — there is no legal prohibition on publishing AI-written content. Disclosure requirements vary by platform and context (academic publishers, major media outlets, and some legal and regulatory contexts require disclosure). What you may lack is the copyright protection that would prevent others from reprinting it.
- Can I use AI-generated code in a commercial product? Generally yes, but check the terms of service of the specific tool. GitHub Copilot, Claude, and most major AI coding tools have terms that allow commercial use of generated code. The copyright status of the code is less clear — if the AI reproduces a substantial portion of copyrighted code from its training data, there could be infringement liability on the output side.
- Can I register AI-assisted creative work with the Copyright Office? Yes, for the human-authored portions. The Copyright Office has accepted registrations for works that include AI-generated elements when the applicant discloses the AI's contribution and demonstrates sufficient human authorship in the overall work.
- Does the disclosure requirement apply? No federal law currently requires disclosure of AI use in commercial content (as of March 2026). Some state laws, platform terms, and professional ethical codes impose disclosure requirements in specific contexts. The EU AI Act has disclosure requirements for AI-generated content that apply to EU consumers.
The Training Data Question: The Bigger Legal Battle
The ownership-of-output question is, in some ways, secondary to a larger legal battle: whether AI companies had the right to train their models on copyrighted works without permission or compensation. Multiple ongoing lawsuits — from the New York Times, from book authors including George R.R. Martin and John Grisham, from visual artists — argue that using copyrighted works for training is infringement. AI companies generally argue this use is protected by fair use doctrine. The resolution of this question will determine whether AI companies owe compensation to content creators for training data and will shape the future of how AI models are trained and licensed.
Pro Tip: If you are building a business on AI-generated content: document your creative process. Keep records of your prompts, your selection decisions, your editorial changes, and the human judgment calls that shaped the final product. If a copyright question arises, the evidence of meaningful human authorship — the selection, arrangement, and creative decisions you made — will define your legal position. Process documentation is not exciting, but it is your insurance policy in a legal landscape that is still being written.