Who owns AI-generated content? What the law and the terms actually say

Two separate systems decide who owns AI-generated content, and most confusion comes from mixing them up. The tool's terms of service decide what rights the platform hands you; copyright law decides whether there is anything to own at all. In the United States, output with no human author gets no copyright — it belongs to nobody, including you. The EU has no specific rule yet but leans the same human-centric way, so the practical question for a creator is not "do I own this?" but "how much of a human hand is in this, and what did I agree to?" This guide maps both layers with sources and dates. It is general information, not legal advice — for a decision with real money attached, talk to a lawyer in your jurisdiction.

Ownership is two questions, not one

When people ask who owns AI content, they are usually asking two different things at once.

Question one: what does the platform grant me? Every AI tool's terms of service say something about your inputs and the outputs. This is contract law. It varies tool by tool, it can change with a terms update, and it binds you the moment you click through.

Question two: does copyright protect the output against everyone else? This is national law, and no platform's terms can change it. A tool can assign you "all its rights" in an output, but if the law says a purely machine-generated image has no copyright, the tool assigned you nothing — there was nothing to give. That is not a trick; it is the honest fine print behind almost every "you own your generations" claim in the industry, and the better-written terms admit it openly.

Keep the two layers separate and the rest of this topic gets much simpler.

The US rule: no human author, no copyright

The United States has the clearest position of any major jurisdiction, and it firmed up recently.

The US Copyright Office published Part 2 of its Copyright and Artificial Intelligence report in January 2025. Its conclusions: human authorship is a bedrock requirement, so a work generated entirely by AI cannot be copyrighted; where a work mixes human and AI contributions, only the human contributions are protectable, judged case by case; and using AI as a tool inside a human creative process does not poison the whole work. The Office also said prompts alone — even long, carefully iterated ones — generally do not amount to authorship, because the machine, not the prompter, determines the expressive details of the result.

The courts agree. In March 2025 the D.C. Circuit ruled in Thaler v. Perlmutter that the Copyright Act requires a human author as a matter of statutory law, upholding the refusal to register an image whose listed author was a machine. The Supreme Court declined to review that decision in March 2026, so it stands. Importantly, the court was explicit that this does not bar copyright on work made with AI — the author just has to be the human, not the machine.

What that means in practice: the raw image or clip that falls out of a generator is, on its own, probably public-domain material in the US. What you build from it — your selection and arrangement of generations, your edits, your composite, the video you cut from AI clips with your pacing and structure — can carry your copyright, because those are your creative choices. The protection attaches to the human layer, and the thicker that layer is, the more you own.

The EU: no specific rule, the same human-centric pull

The EU has no statute that answers the question directly. A European Parliament research briefing from December 2025, Copyright of AI-generated works: approaches in the EU and beyond, states it plainly: the EU lacks specific rules on the copyrightability of AI-generated works, and existing case law points toward requiring meaningful human involvement. EU copyright protects works that are the author's own intellectual creation — a standard built around human choices, which purely machine-generated output struggles to meet for the same reason it fails in the US.

Two differences are worth knowing. First, the UK — no longer in the EU, but often lumped in — has an unusual provision that grants copyright in "computer-generated" works to the person who made the arrangements for their creation, with no human-authorship requirement. It is a genuine outlier, its scope is debated, and it should not be generalized to the continent. Second, the EU AI Act does regulate AI content, but on a different axis: it imposes transparency and marking obligations, not ownership rules. Whether you must disclose that content is AI-generated is a separate question from whether you own it — our EU AI Act labeling guide for creators covers that side.

The honest summary for an EU creator: your position is less settled than an American's, but the safe working assumption is identical — the human contribution is what the law protects, so make yours visible.

What the terms of service actually give you

Now the contract layer. Read the ownership clause of any tool you rely on, because the spread across the industry is wide. Two well-known examples, as of August 2026:

Some large AI providers assign outputs outright. OpenAI's terms of use say the company "assigns to you all our right, title, and interest, if any, in and to Output" — note the load-bearing "if any," which is the copyright reality from the previous sections written into the contract. Midjourney's terms of service grant paid subscribers ownership of their assets "to the fullest extent possible under applicable law," while users without a subscription get only a limited license to their images rather than ownership — and all users grant Midjourney a broad, perpetual license back to the content they make on the platform.

Three things to check in any tool's terms:

1. Do outputs come to you as ownership, an assignment, or a license? And does the grant depend on being on a paid plan?
2. What license do you give back? A perpetual right for the platform to reuse and remix your content matters if your work is commercial or client-facing.
3. Is exclusivity promised? It almost never is. Generative models can produce similar or identical output for different users from similar prompts, and no terms document can stop that.

For completeness, PrismPoster's own terms take the pass-through approach: you own your inputs, and for generated outputs we pass through the rights the underlying AI provider grants, to the extent the law allows — with no pretense that a terms page can conjure copyright where the law grants none. We would rather say that plainly than print "you own everything" and let the fine print take it back.

Where copyright is not the only layer

Two kinds of content carry rights that sit entirely outside the copyright question, and both catch creators off guard.

Music. A generated track involves the composition, the recording, and the platform's license terms — and commercial use, monetization, and Content ID interact with all three. If music is part of your workflow, AI music licensing explained walks through that stack; the short version is that "who owns the song" and "what am I allowed to do with the song" get different answers more often than you would expect.

People. A real person's face and voice are protected by likeness and personality rights, which exist regardless of who owns the copyright in the media. Owning an AI-generated video does not make it legal to put a recognizable real person in it without consent — and in a growing list of jurisdictions, the penalties for that have teeth. This is why PrismPoster's image-to-video will not animate a raw photo of an identifiable real person. Fictional characters, artwork, products, and landscapes animate freely.

Practical guidance: how to actually protect your work

Given all of the above, here is what a working creator can do this week.

Put a human layer on anything that matters. A raw single generation is the weakest position you can hold. Selecting from dozens of generations, compositing, editing, color work, cutting clips into a sequence with your structure and timing — that is authorship the US Copyright Office has said it will recognize, case by case. Treat generation as raw material, not as the finished work.

Keep your process. Save prompts, iterations, drafts, and project files. If you ever need to show which parts of a work are yours — to a registrar, a client, or a platform dispute process — a documented trail of human decisions is the evidence.

Read the ownership clause of every tool in your stack. Ten minutes per tool. Check the grant, the license back, and whether your plan tier changes either. If you cannot find a clear answer on the terms page, that is an answer.

Be straight with clients. If you deliver AI-heavy work, say so in the contract and be precise about what rights you can actually convey. Promising exclusive copyright in a raw generation is promising something you may not have.

Keep provenance intact. PrismPoster's paid exports carry no visible watermark, but C2PA content credentials are embedded — a tamper-evident record of how the file was made. As disclosure rules like the EU AI Act come into force, a clean provenance trail is an asset, not a liability: it proves what you made and how, on your terms.

Frequently Asked Questions

Can you copyright AI-generated art?

Not the raw output by itself, in the US — a work with no human author cannot be registered, and prompts alone generally do not count as authorship. What you can protect is your human contribution: the selection, arrangement, and editing you add on top. The more creative decisions you make after generation, the stronger your position.

Who owns an image I generate with an AI tool?

Check two layers. The tool's terms decide what the platform grants you — some assign outputs to you, some reserve ownership for paid tiers, as of August 2026. Copyright law then decides whether that grant contains anything: a purely machine-generated image may not be protectable at all, in which case nobody owns it, and someone else could legally use an identical output.

Is AI content ownership different in the EU?

The EU has no specific statute on AI-generated works yet, per a December 2025 European Parliament briefing, but its copyright standard is built on human intellectual creation, which pulls the same direction as the US rule. The EU AI Act adds transparency obligations for AI content but says nothing about who owns it. Practically, EU creators should work the same way: document the human contribution.

Does PrismPoster claim ownership of what I make?

No. You own your inputs, and generated outputs come to you with the rights the underlying AI provider grants passed through, to the extent the law allows — our terms say exactly that rather than overpromising. Paid exports have no visible watermark, free exports carry an AI-disclosure label, and embedded C2PA credentials record provenance without asserting any claim over your work.

Can someone steal my AI-generated content?

If the content has no copyright — a raw generation with no human authorship — then legally there may be nothing to steal, which is exactly why the practical advice above centers on adding and documenting a human layer. Your edited, arranged, composited work is protectable; your untouched generation mostly is not. Contracts, trademarks, and likeness rights can add protection that copyright does not.

Where this leaves you today

You cannot change the law, but you can change how much of your work sits on the protectable side of the line, and that is a workflow choice. The way to make AI output defensibly yours is to work it: select, edit, arrange, cut. A free PrismPoster account comes with a one-time grant of 200 credits — no card — which is enough to generate images and video clips today and practice exactly that: generate raw material, then shape it into something with your decisions in it. The pricing page has the full cost table if you want to see the numbers first, and AI credits explained covers how the metering works. Whatever tool you use — ours or anyone's — read the terms, keep your drafts, and put your hands on the work.