TL;DR: Who owns AI-generated content? Two separate rules answer that. The tool's terms of service decide who holds the rights between you and the provider, and most major tools hand them to you. Copyright law decides whether there are any rights to hold, and in the US a purely AI-generated image, clip or paragraph has no copyright owner at all, because copyright needs a human author. So you can usually use and sell what you generate, but you often can't stop anyone else from copying it, unless you add enough of your own creative work.
A designer generates a product shot, a marketer publishes an AI video ad, a founder ships a blog written mostly by a chatbot. Then the same question comes up: is this ours?
The honest answer is "partly." You almost always have the right to use what you generate. Whether you own it in the legal sense, with the power to stop competitors from copying it, depends on how much of it came from you. That gap catches out a lot of creators and businesses.
This guide explains where the law stands as of October 2026 in the US, UK, EU and China, what the big tools' terms actually say, and how to set up your work so the parts that matter are protected.
This article is general information, not legal advice. If a specific asset is valuable to your business, talk to an intellectual property lawyer.
Key Takeaways
- Ownership has two layers. Contract (the tool's terms) and copyright (the law). A tool can give you "all rights" and still leave you with no copyright, because there may be none to give.
- US law requires a human author. The Supreme Court declined to hear Thaler v. Perlmutter in March 2026, leaving that rule settled for now.
- Prompts alone don't make you the author under current Copyright Office guidance, however detailed they are.
- Your human contribution is protectable: your own text, art and footage, real edits, and creative selection and arrangement of AI material.
- Using is not the same as owning. Most paid plans let you use outputs commercially, but you carry the risk if an output copies someone else's work.
Two different questions: who holds the rights, and do any rights exist?
When people ask who owns AI content, they are usually mixing two questions.
1. Between you and the AI company, who gets the output? This is a contract question. The terms of service you accepted decide it. OpenAI's terms, for example, say you own the output and that OpenAI assigns you "all our right, title, and interest, if any" in it. Note the words "if any." The company is handing over whatever rights it has, without promising that those rights exist.
2. Does the output have a copyright at all? This is a legal question, and no contract can change the answer. If the law says a work has no human author, it has no copyright, and nobody owns it in the copyright sense, not you, not the AI company.
Keep those two layers apart and most of the confusion goes away. You can have full contractual rights to use an image and still have no way to stop a competitor who copies it.
Is AI-generated content copyrighted? The US position
The human authorship rule
The US Copyright Office has refused for years to register works made without a human author, and the courts have backed it.
- Thaler v. Perlmutter. Stephen Thaler tried to register an image, A Recent Entrance to Paradise, listing his AI system as its sole author. The district court and then the D.C. Circuit (March 2025) held that the Copyright Act requires a human author. On March 2, 2026, the Supreme Court denied certiorari, so that ruling stands.
- The Copyright Office's AI report, Part 2 (January 2025). The Office concluded that outputs are protected only where a human has determined enough of the expressive elements. Its key line for everyday users: prompts alone are not enough, because the person prompting does not control how the system turns the prompt into an image or text.
What is protected when you use AI
The Office was just as clear that AI-assisted work is not automatically unprotected. Your human contribution can be protected when it is visible in the final work:
| Your contribution | Protected? | Example |
|---|---|---|
| Prompt only, output used as-is | No | A Midjourney image published straight from the grid |
| Your own original material, transformed by AI | Your original material, yes | Your sketch or photo used as the base for an AI render |
| Substantial human edits to an output | The edits, yes | Repainting, recomposing or rewriting large parts by hand |
| Creative selection and arrangement | The arrangement, yes | A comic, a video edit or a layout built from many AI clips |
| AI-assisted tools inside normal editing | Generally yes | Using AI fill or upscaling inside a work you authored |
The best-known example is Zarya of the Dawn (2023). Kris Kashtanova wrote a graphic novel illustrated with Midjourney images. The Copyright Office protected the text and the selection and arrangement of the images, but not the individual images themselves.
The open test case is Allen v. Perlmutter in federal court in Colorado. Jason Allen used more than 600 prompts, plus edits, to create Théâtre D'opéra Spatial, which won a prize at the 2022 Colorado State Fair. The Copyright Office refused to register the AI-generated parts, and summary judgment motions were still pending as of late summer 2026. That case may say how much iteration and editing it takes to count as authorship.
Registering a work that includes AI
If you register a work with the Copyright Office, you must disclose AI-generated material that is more than minimal and claim only the human-authored parts. Leaving it out can put the registration at risk later, which matters if you ever need to sue.
What the major AI tools' terms say
Because the copyright layer is often thin, the contract layer matters a lot in practice. Here is how some widely used tools allocate rights. Terms change often, so always check the current version for your plan.
| Tool | Who holds rights in outputs | Notable conditions |
|---|---|---|
| ChatGPT / OpenAI | You own the output; OpenAI assigns its rights "if any" | Others may receive similar outputs; you're responsible for how you use them |
| Midjourney | You own your assets | Companies (or their employees) with over $1M in yearly gross revenue need a Pro or Mega plan to own outputs |
| Adobe Firefly | Usable commercially | Trained on licensed and public-domain content; IP indemnity only on certain enterprise licenses |
| AI video and image studios | Usually the user, on paid plans | Free tiers often add watermarks or restrict commercial use |
Most creator tools follow the same pattern: free tiers for testing, paid tiers with commercial rights and watermark-free exports. That holds for image tools and for any AI video generator you use for client work or ads. Whatever the tool, read the license on the plan you actually use, not the one on the pricing page.
Similar outputs. Many terms point out that other users may get outputs similar to yours. Combined with the copyright gap, that means you usually can't claim exclusivity over a raw AI image or paragraph.
Can you use AI-generated images commercially?
Usually yes, and most of the time the risk isn't about who owns the output. It's about whether the output infringes something that someone else owns. Before you put AI content in an ad, on a product or on a client's site, run these checks:
- Plan and license. Does your plan allow commercial use? Is there a revenue threshold or a seat limit?
- Recognizable works. Does the image reproduce a famous artwork, a film still, a character or a specific photographer's shot? Copying protected expression is infringement whether a human or a model did it.
- Trademarks and logos. Brand names, logos and packaging in an output can create trademark problems, especially in ads.
- Real people. A realistic face or voice that looks or sounds like a real person raises right-of-publicity and deepfake issues. See our guides on deepfake laws and on whether AI voice cloning is legal.
- Disclosure. Sponsored posts and some ads must be clearly labeled, and AI-generated people or endorsements can be misleading. Our guide to FTC disclosure rules for influencers covers this.
- Client contracts. If you deliver work to clients, your contract should say that AI tools may be used and what rights you can actually transfer.
Outside the US: UK, EU and China
The question gets different answers in different countries, which matters if you sell or publish internationally.
| Jurisdiction | Current position (October 2026) |
|---|---|
| United States | Human authorship required; AI-assisted works protected only for the human contribution |
| United Kingdom | Section 9(3) of the CDPA gives "computer-generated works" with no human author a 50-year copyright, owned by whoever made the arrangements for the work. The government's March 2026 report said its preferred option is to repeal this rule |
| European Union | Copyright requires the "author's own intellectual creation," which points to human creative choices. The AI Act adds transparency duties for AI-generated content, phasing in from August 2026 |
| China | The Beijing Internet Court (Li v. Liu, 2023) and the Changshu People's Court (2025) protected AI images where the user showed real creative input through prompts, parameters and editing |
The UK rule is unusual: it is one of the few laws that expressly protects works with no human author. If the UK repeals it as planned, the major Western jurisdictions will line up behind the human-authorship approach.
What about the training data lawsuits?
A separate set of cases asks whether training AI models on copyrighted works is lawful. They don't decide who owns your outputs, but they shape the risk around the tools you use.
- Thomson Reuters v. Ross Intelligence. On September 29, 2026, the Third Circuit held that copying Westlaw headnotes to train a competing legal research tool was not fair use. It was the first federal appeals decision on AI training, though it involved a non-generative tool.
- Bartz v. Anthropic. Authors' claims over books downloaded from pirate sites ended in a $1.5 billion settlement, which received final court approval in July 2026.
- Getty Images v. Stability AI (UK). In November 2025 the High Court rejected Getty's remaining secondary copyright claim, finding the model weights were not "infringing copies." Getty was granted permission to appeal.
For most businesses, the practical takeaway is to favor tools that are open about their training data or that offer indemnity, especially for high-visibility campaigns.
How to protect the content you make with AI
You can't make a raw output copyrightable by wishing it so, but you can build workflows where the valuable parts belong to you.
- Add real human authorship. Write the script yourself, edit the footage, retouch and recompose images, and arrange outputs into something new. The more your choices show in the result, the more there is to protect.
- Keep a record. Save drafts, prompts, source files and edit history. If ownership is ever disputed, you'll need to show what you did.
- Use contracts. Clients, freelancers and agencies should agree in writing on who owns AI-assisted deliverables and who carries the infringement risk.
- Use other protections. Trademarks protect brand names and logos regardless of who drew them, and trade secret law can protect prompt libraries and internal workflows.
- Check before you publish. Run important assets through a reverse image search and avoid prompts that name living artists, brands or real people.
If you need to check how courts have treated a specific rights question, AI case law search lets you find and read the actual decisions, and our roundup of the best AI tools for IP lawyers covers tools built for deeper IP work.
Get cited answers on your own rights questions
AI-content law is moving fast, and a blog post can only cover the general rules. When you have a specific question, such as whether a client contract covers AI-generated deliverables or how a recent decision affects your use case, the LegesGPT legal AI chatbot gives you answers grounded in statutes and case law, with citations and links to the sources so you can check every point yourself. It can also review your contracts and flag clauses about IP ownership and AI use before you sign.
Frequently Asked Questions
Who owns AI-generated content?
Two different things decide it. The tool's terms of service decide who holds whatever rights exist between you and the provider, and most major tools such as ChatGPT, Midjourney on the right plan, and Adobe Firefly give those rights to the user. Copyright law decides whether any rights exist at all, and in the US a purely machine-made output has no copyright owner because the law requires a human author. In practice you can usually use the output, but you may not be able to stop others from copying it.
Is AI-generated content copyrighted?
Not by default in the United States. The Copyright Office and the courts, most recently the D.C. Circuit in Thaler v. Perlmutter (the Supreme Court declined review in March 2026), hold that copyright requires human authorship. The human parts of a work, such as your own text, your edits, and your creative selection and arrangement of AI material, can still be protected.
Can you use AI-generated images commercially?
Usually yes, if the tool's terms allow it on your plan. Check three things before you publish: that your plan grants commercial rights (Midjourney, for example, requires a Pro or Mega plan for companies with more than one million dollars in annual revenue), that the image does not copy a recognizable artwork, character, logo or real person, and that you are comfortable without exclusive rights, since others may generate similar images.
Are prompts enough to make me the author?
Not under current US guidance. The Copyright Office's January 2025 report concluded that prompts alone do not give a user enough control over the output to make them its author, even long or repeated prompts. What can count is expressive input you can see in the result, such as your own artwork used as a base, substantial edits, or a creative arrangement of several outputs.
Can I register a work that includes AI-generated material?
Yes, but you must disclose the AI-generated parts that are more than minimal and claim only the human contribution. The comic book Zarya of the Dawn is the standard example: the Copyright Office protected the author's text and the arrangement of images, but not the individual Midjourney images.
Is the law the same outside the US?
No. The UK has a special rule that gives computer-generated works a 50-year copyright owned by whoever made the arrangements for the work, although the government said in March 2026 that it prefers to repeal it. EU courts require the author's own intellectual creation, which points to human input. Chinese courts in Beijing (2023) and Changshu (2025) have protected AI images where the user showed real creative choices.
Can I get sued for using AI-generated content?
Yes, if the output infringes someone else's rights. If an image closely resembles a copyrighted work, uses a trademark, or shows a real person's likeness without permission, the person publishing it can be liable even though an AI made it. Some enterprise plans, such as certain Adobe Firefly licenses, include IP indemnity, but most consumer plans do not.
How do I protect content I make with AI?
Keep a record of your process, such as drafts, prompts, edits and source files, so you can show human authorship. Add meaningful human contribution rather than publishing raw outputs. Use contracts with clients and freelancers that state who owns AI-assisted deliverables, and rely on trademarks and trade secrets for brand assets that copyright may not cover.
