Who Owns AI-Generated Content: What the Law Says Before You Sell It
Ask who owns AI-generated content and the tool you used will tell you, cheerfully, that you do. OpenAI’s terms say you own the output. Midjourney’s say you own the assets. Case closed, apparently.
Then read the rest of the sentence. OpenAI assigns you “all our right, title, and interest, if any.” Midjourney grants ownership “to the fullest extent possible under applicable law.”
Both companies are being careful about something most people never notice. A company can hand you whatever rights it holds. It cannot hand you a right that the law says does not exist — and in the United States, the law on this is now settled in a way that matters to anyone selling an image, a design, a template, a book, or a course that a model helped produce.
This article goes through what was actually decided, what it means for different kinds of businesses, and where it leaves you if you are already selling AI-assisted work.
Two Questions That Keep Getting Confused
Almost all the confusion here comes from collapsing two separate questions into one.
Can you use it? This is a contract question, answered by the terms of service of whatever tool you used. Usually the answer is yes, including commercially, sometimes with conditions.
Can you stop someone else from using it? This is a copyright question, answered by law, not by the tool. It determines whether you have anything to enforce when a competitor copies your work.
You can comfortably have the first without the second. That combination — free to sell, powerless to protect — is the actual position a lot of people are in without knowing it.
What the Copyright Office Actually Decided
In January 2025 the U.S. Copyright Office published the second part of its report on copyright and artificial intelligence, dealing specifically with copyrightability. It is the clearest official statement available, and it is unusually direct.
On prompts, the report concludes that prompts alone do not provide sufficient human control to make users of an AI system the authors of the output. Prompts, it says, “essentially function as instructions that convey unprotectible ideas.”
That sentence is the one to sit with. Copyright has never protected ideas — only the specific expression of them. A prompt describes what you want. The model produces the expression. Writing a longer, more detailed, more clever prompt does not change which side of that line you are standing on.
The Office is equally clear about the boundary: “Copyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements.”
It also concluded that no new legislation is needed, since “questions of copyrightability and AI can be resolved pursuant to existing law.” Which means this is not an interim position waiting to be replaced. It is the position.
The Courts Settled It Too
A report is guidance. A court ruling is law, and there is one.
In Thaler v. Perlmutter, the D.C. Circuit held in March 2025 that human authorship is required as a matter of statutory law. A work generated autonomously by a machine, with no human author, cannot be registered.
The Supreme Court declined to review that decision in March 2026, leaving it standing.
For practical purposes, that closes the question in the United States. Anyone still telling you the law is unsettled on whether a machine can be an author is working from a 2023 understanding.
| Protected | Not protected | |
|---|---|---|
| Output from a prompt alone | ✗ | |
| Your own writing or artwork visible in the output | ✓ | |
| Your creative selection and arrangement of AI material | ✓ (the arrangement) | |
| Heavy human editing and reworking of AI output | ✓ (your contribution) | |
| A detailed, long, expert prompt | ✗ | |
| A machine-generated work with no human author | ✗ |
Who Owns AI-Generated Content Once You Add Real Work
This is the part that gets lost in headlines, and it is where the practical answer lives.
The Office did not say that anything touched by AI is unprotectable. It said the opposite, twice over.
First, your own material survives: “where a human inputs their own copyrightable work and that work is perceptible in the output, they will be the author of at least that portion of the output.” Feed in your own photograph, your own text, your own drawing, and what remains visible of it is still yours.
Second, arrangement counts. A human may “select or arrange AI-generated material in a sufficiently creative way that the resulting work as a whole constitutes an original work of authorship.”
So the question is not whether you used AI. It is how much of the finished thing is a decision you made rather than a decision the model made.
A hundred generated images dumped into a pack is not protectable. Twenty generated images you selected from six hundred, sequenced deliberately, edited, and assembled into a coherent collection — the selection and arrangement is yours, even though no single image is.
That distinction is the one worth building a working method around.
Why “You Own the Output” Means Less Than It Sounds
Now back to the terms of service, with the law in hand.
OpenAI’s wording is precise: “As between you and OpenAI, and to the extent permitted by applicable law, you (a) retain your ownership rights in Input and (b) own the Output. We hereby assign to you all our right, title, and interest, if any, in and to Output.”
Read “as between you and OpenAI.” That clause settles a dispute between two parties — you and the company. It says nothing about a third party who copies your work.
Read “if any.” OpenAI is assigning whatever rights it has. If the law says there is no copyright in that output, then the thing being assigned to you is nothing at all, and the sentence is still perfectly true.
Midjourney’s terms do the same work with different words: “You own all Assets You create with the Services to the fullest extent possible under applicable law.” The limit is right there in the sentence.
Neither company is being deceptive. They are describing a contractual position accurately. The misreading happens on our side, because we read “you own the output” and stop before the qualifier that defines it.
The Clauses People Miss
Beyond ownership, the terms carry conditions that catch specific businesses off guard.
Midjourney’s revenue threshold. “If you are a company or any employee of a company with more than $1,000,000 USD a year in revenue, you must be subscribed to a ‘Pro’ or ‘Mega’ plan to own Your Assets.” Ownership here is tied to the plan you are paying for — a detail with real consequences for anyone whose employer crosses that line.
Upscaling other people’s work. Midjourney states plainly: “If you upscale the images of others, these images remain owned by the original creators.” Touching someone else’s asset does not make it yours.
The licence running the other way. Midjourney takes “a perpetual, worldwide, non-exclusive, sublicensable no-charge, royalty-free, irrevocable copyright license” to content you input. Whatever you upload, you are granting broad rights over.
Output is not unique. OpenAI notes that “output may not be unique and other users may receive similar output from our Services,” and that its assignment “does not extend to other users’ output.” Two people can receive near-identical results, and neither has a claim against the other.
That last point deserves emphasis, because it is the one that bites commercially. The thing you are selling may already exist in someone else’s catalogue, produced independently, with no wrongdoing by anyone.
What This Means by Type of Business
The exposure varies enormously depending on what you actually sell.
| Business | Real exposure |
|---|---|
| Freelance writing or design, delivered to a client | Low — the client is buying a solved problem, not a copyright they will enforce |
| Service work with a contract promising original IP | High — you may be promising something you cannot deliver; read the clause |
| Stock-style assets sold at volume | High — the whole product is unprotectable and trivially duplicated |
| Templates, worksheets, planners | Medium — the individual elements may be unprotectable, your arrangement may not be |
| Books and courses | Low to medium — your writing and structure are yours; purely generated sections are not |
| A brand logo or identity | High — the one asset you most need to protect is the worst candidate for pure generation |
| Content on your own site | Low — you are monetising attention, not licensing the text |
Look at the pattern. Exposure rises exactly where the product is the asset and falls where the asset is a by-product of a service. Someone selling a digital product to an audience they do not yet have is in the higher-risk column by definition, because the file is the entire thing being bought.
The lower-risk column is not an accident either. It lines up closely with the categories of AI-assisted work clients actually pay for, where what is being bought is a problem removed rather than an asset licensed.
The Contract Clause Worth Checking Tonight
If you do freelance or agency work, there is one concrete action in this whole article that is worth doing immediately.
Open your client contract or the platform agreement you work under, and find the intellectual property clause. A standard one assigns “all right, title and interest” in the deliverable to the client, and frequently includes a warranty that the work is original and that you have the right to assign it.
If a meaningful part of the deliverable was generated rather than authored, that warranty is a promise you may not be able to keep. Not because you did anything wrong, but because there may be nothing there to assign.
This is not a reason to stop using these tools. It is a reason to know what you are signing, and to raise it before a dispute rather than during one.
Where Disclosure Fits
Ownership and disclosure are separate obligations, and it is worth not confusing them either.
Nothing in copyright law requires you to tell a buyer how a work was made. Specific contracts, platforms, and publishers do, and those requirements are contractual rather than legal.
Search engines sit in a third category again. Google’s documentation does not penalise AI assistance as such — what its policies actually target is different, and conflating “will this rank” with “do I own this” leads people to optimise for the wrong risk entirely.
A Working Position
Five things follow from the above, in rough order of how much they matter.
Stop treating the prompt as the work. The Office was explicit that prompts convey unprotectible ideas. Effort spent perfecting a prompt builds no ownership. Effort spent on selection, editing, and arrangement does.
Put your own material in. Your photograph, your text, your drawing, your recording. Whatever of it remains perceptible in the output is authored by you, and that is the most reliable route to having something protectable at all.
Keep your working files. Drafts, versions, edits, the sequence of decisions. If you ever need to show where the human contribution sits, that record is the evidence, and it only exists if you kept it.
Never build a brand asset on pure generation. A logo, a mascot, a character, a name — these are the assets whose entire value is the ability to stop others using them. They are the worst possible candidates.
Read the IP clause before signing, not after. Covered above, and it is the one with a counterparty who may eventually care.
What This Article Does Not Settle
Three honest limits.
This is United States law. The Copyright Office and the D.C. Circuit govern the U.S. The UK, EU, Brazil, and others are each working through this separately and have not all landed in the same place. If you are selling into a particular market, that market’s rules are the ones that apply.
The training-data question is wide open. Whether models were lawfully trained on copyrighted material is a separate fight, with active litigation and no settled answer. Nothing here speaks to it.
This is not legal advice. It is a summary of published official sources, written by someone who is not a lawyer. If real money depends on the answer in your situation, the cost of an hour with an IP lawyer is small next to the cost of finding out later.
Frequently Asked Questions
Can I legally sell something made with AI? Generally yes, provided the tool’s terms permit commercial use — and the major ones do. Selling it and being able to stop others copying it are separate questions, and the second one usually has a less comfortable answer.
Can I copyright an image generated from a prompt? Not the image itself, under current U.S. guidance. The Copyright Office concluded that prompts alone do not give the user sufficient control over the expressive elements, so the raw output is not protected.
What if my prompt was extremely detailed? Length and detail do not change the analysis. The report describes prompts as instructions conveying unprotectible ideas, and copyright has never protected ideas regardless of how precisely they are stated.
So nothing involving AI can be copyrighted? No — that is the common misreading. Your own material that remains perceptible in the output is yours, and a sufficiently creative selection and arrangement of AI-generated material can be protected as a work in its own right.
Does OpenAI’s “you own the output” clause give me copyright? It gives you whatever rights OpenAI holds, assigned “if any,” and settles the position between you and OpenAI. It cannot create a copyright that the law says does not exist, and it does not bind third parties.
Can two people end up selling the same AI-generated asset? Yes. OpenAI’s terms state output may not be unique and that other users may receive similar output. With no copyright in the raw output, neither party generally has a claim against the other.
I run a company — does any of this change? Possibly. Midjourney requires companies with over $1 million in annual revenue, and their employees, to be on a Pro or Mega plan to own their assets. Check the terms of each tool your team uses against your own situation.
Should I tell clients I used AI? Copyright law does not require it. Your contract might, and some platforms and publishers do. Check what you actually signed, and be aware that a warranty of originality is a separate promise from a disclosure obligation.
Conclusion
The question of who owns AI-generated content has a clearer answer in the United States than most coverage suggests, and it is not the answer the tools’ marketing implies.
Purely generated material has no copyright owner. The terms of service that appear to give you ownership are, read carefully, assigning you whatever rights exist — which in that case is nothing. The Copyright Office said so, the D.C. Circuit said so, and in March 2026 the Supreme Court declined to disturb it.
What that leaves is more useful than it first appears. Protection attaches to the human contribution: your own material carried into the work, your editing, your selection, your arrangement. The more of the finished thing represents a decision you made rather than one the model made, the more there is to own.
Which turns out to be the same conclusion the labour market reaches by an entirely different route. The work that holds its value is the work that requires a decision — and copyright, it turns out, draws its line in almost exactly the same place. The parts of your work that are just generation are the parts nobody owns and anybody can reproduce. The parts that are judgment are the parts that are yours.
A Note from Lucas
I went into this expecting a grey area and found something closer to a straight line, which almost never happens with a legal question. What actually unsettled me was the phrase “if any” in OpenAI’s terms — I had read that clause before, more than once, and never registered it. It is doing enormous work in a sentence most of us skim. The practical change I made for myself after writing this is small and probably the right one for most people: I keep the drafts now, every version, because the record of what I decided is the only evidence that the human part existed. If you sell anything where the file is the product, I would read your own tool’s terms tonight rather than take my summary for it.
About the Author
Lucas Palma is the founder and editor of Info Is The Key. He writes about online business, entrepreneurship, and the practical mechanics of building income independently, working from primary sources and official data rather than recycled advice.
Last Updated: October 2026 | Reading Time: 11 minutes
Editorial note: This article summarises published U.S. government sources, court decisions, and company terms of service as of October 2026. It is educational and does not constitute legal advice, and its author is not a lawyer. Copyright law differs by country, terms of service change without notice, and litigation on related questions is ongoing. Verify current terms directly with each provider and consult a qualified intellectual property lawyer regarding your own circumstances.
