Your AI tool's terms probably say you own the output. That's true and it doesn't mean what most people think — a contract can assign whatever rights the platform holds, but it cannot create a copyright the law never granted.
General information about a developing legal area, not legal advice. Positions vary by jurisdiction and by facts — take advice on anything consequential.
What's now settled
The US human-authorship question closed in early 2026, and the sequence is worth stating precisely because it's frequently reported wrong.
In Thaler v. Perlmutter, the D.C. Circuit held in March 2025 that the Copyright Act requires eligible work to be "authored in the first instance by a human being." Rehearing was denied in May 2025, and on 2 March 2026 the Supreme Court declined to hear an appeal.
That last step matters and it's widely mis-described. Several sources refer to a "March 2026 Supreme Court ruling." There wasn't one — the Court denied certiorari, meaning it chose not to take the case, which leaves the lower court's holding standing without the Supreme Court having endorsed its reasoning. The practical effect is similar; the precision isn't merely pedantic, since a denial of certiorari carries different weight from a decision on the merits.
The part most coverage under-reports is the court's clarification. The human authorship requirement does not prohibit copyrighting work made with AI assistance — it requires only that the author be the human who created, operated or used the system, rather than the machine.
The distinction that matters "AI can't hold copyright" is settled. "AI-assisted work can't be copyrighted" is false, and confusing the two produces both false comfort and unnecessary caution.
The contract trap
The single most useful correction for marketers, and it catches people out precisely because the terms genuinely do say what they think.
Platform terms of service commonly assign output ownership to the user. What that actually means is that the platform isn't claiming rights against you — a useful and real assurance. It does not mean a protectable copyright exists.
If the output has no human author, there is no copyright for anyone to transfer. Two parties cannot manufacture one by agreement. So "we own it, it says so in the terms" answers a question about the platform's claims and says nothing about whether you can stop a competitor reproducing it.
The risk marketers actually face
Most coverage frames this as infringement exposure. For a marketer publishing AI-assisted content, the more common practical issue runs the other way.
You may be unable to stop anyone copying it. If a piece of output has no human author, it may have no copyright owner — which means a competitor could reproduce it without infringing anything.
For most marketing content, that's a minor concern. A blog post's value lies in being found and read, not in being exclusive. But it matters considerably for a specific set of assets:
- Brand imagery and visual identity intended to be distinctive and defensible
- Campaign creative you'd expect to enforce against imitation
- Anything licensed to third parties, where you're granting rights you may not hold
- Assets underpinning a commercial position — characters, mascots, signature formats
For those, the protectability question is a business question rather than a compliance one, and it's worth resolving before the asset becomes load-bearing.
The separate question — liability for what models were trained on — sits primarily with providers rather than users in most circumstances, and is the subject of active litigation. Several major cases were moving toward appellate decisions expected across 2026 and 2027, so anything stated confidently about fair use in AI training is currently a prediction.
Adcreative
AI platform that generates conversion-focused ad creatives, banners and social posts in seconds.
Best for: Marketers generating high-converting ad creatives fast with AI
What is protectable
The constructive half, and there's more here than the headlines suggest.
The US Copyright Office has reported registering hundreds of works incorporating AI-generated material, with protection covering the human author's contribution. The registration practice is established rather than theoretical.
The pattern comes from the Zarya of the Dawn decision, where a graphic novel used generated images. The Office found the images themselves unprotectable, while the human-written text and the selection, coordination and arrangement of the elements remained protected.
So what earns protection:
- Human-authored elements — text you wrote, photographs you took, illustrations you drew.
- Substantial modification of generated output, where a person's creative choices are visible in the result.
- Selection and arrangement — which outputs you chose from many, in what order, combined how. This is a genuine authorial act and it's the most commonly overlooked.
What generally doesn't, on current US practice: the prompt alone, however elaborate. Whether iterative prompting plus post-processing crosses the line is the live question — a case testing prompts-as-authorship has been expected to produce a ruling during 2026, so this is the part of the picture most likely to move.
Jurisdictions diverge, and it matters
The part most coverage skips, and the reason a single global answer doesn't exist.
| Jurisdiction | Position |
|---|---|
| United States | Human authorship required; no explicit computer-generated works provision |
| United Kingdom | Long-standing provision attributing authorship of computer-generated works to the person who made the arrangements for their creation |
| India | Copyright Act provision defining the author of a computer-generated work as the person who "causes the work to be created" |
| China | Courts reportedly more willing to find protection where creativity traces to human input |
The UK and Indian provisions are the interesting ones, because both predate generative AI by decades — they were written for a world of computer-assisted output, not systems that produce finished creative work. Whether they extend cleanly to generative tools is genuinely unresolved.
India illustrates the tension directly. In 2020 the Copyright Office granted registration for an artwork created with an AI application, listing both a human applicant and the tool in the authorship details. It later reconsidered and issued a withdrawal notice seeking clarification of the AI system's legal status. That sequence — registration, then reconsideration — captures the uncertainty better than any position statement would.
Commentary also disagrees about the Indian position, with some arguing the Act does not permit non-human authorship at all and others pointing to the computer-generated works provision as a possible route. That disagreement is itself the current state of affairs.
For anyone publishing across markets, the practical implication is that your protection is only as strong as the weakest jurisdiction you care about — and building around the most restrictive position is the conservative approach.
The separate question of disclosure
Distinct from ownership and moving faster.
The EU AI Act introduced transparency obligations for AI-generated content, alongside requirements for providers to disclose information about training material. Further proposals — a training data register, opt-out rights, remuneration frameworks — were reported moving through the European legislative process during 2026 without being settled.
The direction is toward more disclosure rather than less, which makes "we'll address it later" a weakening position. The practical implications for synthetic media specifically are worked through in the state of AI-generated video.
Worth separating three things that get conflated: whether you own it, whether you must say it's AI-generated, and whether the model was trained lawfully. Those are different questions with different answers in different places, and treating them as one produces confusion in both directions.
What to actually do
Proportionate to risk rather than uniformly cautious.
Sort your assets by whether protectability matters. Most blog posts, social captions and ad variations don't need defending. Brand marks, signature imagery and licensed material do. Apply effort accordingly rather than treating everything the same.
For assets that matter, ensure genuine human authorship — meaningful creative decisions, substantial modification, or human-created elements — and document it at the time.
Don't rely on platform terms as evidence of ownership. Useful assurance about the platform's claims; not a copyright.
Check what you're warranting in client contracts. Agencies in particular should look at whether they're representing that deliverables are original and protectable, since that's a promise AI-assisted work may not support. This is a contract review worth doing once.
Keep the verification discipline separate. Whether output is accurate is a different problem from whether you own it, and it carries its own risks — the ground covered in the AI content QA checklist. Both belong in the same workflow, as described in building an AI content workflow.
Take advice on anything consequential, particularly if you operate across jurisdictions where the positions differ.
A note on the sources
Worth flagging, because this topic attracts confident writing on unsettled questions.
Much of the accessible material describes a denial of certiorari as a ruling, states the Indian position as though it were resolved, or presents fair use in AI training as though appellate courts had decided it. None of those is accurate as of writing.
The genuinely settled points are narrow: a machine cannot be an author under US law, AI-assisted work with human authorship can be registered, and registration practice covers the human contribution. Almost everything else — how much human input suffices, whether training is fair use, how older computer-generated works provisions apply — is in motion, with several decisions expected across 2026 and 2027.
Which is a reason to build practices that survive either outcome. Documenting human contribution is useful whether the line moves toward or away from you, and it costs almost nothing.
It also connects to a broader point about why generic output is commercially weak regardless of the legal position: content that anyone could have produced is content anyone can reproduce, which is the sameness problem expressed as a property question. The assets worth protecting tend to be the ones with genuine human judgement in them — which is also what makes them work.
If your organisation is producing at volume without anyone having looked at which outputs are load-bearing enough to need defending, that gap is worth closing before it becomes a dispute — and it's where a brand and design partner producing original work under clear terms removes the question entirely for the assets where it matters most.
The short version
Human authorship is settled in the US — the Supreme Court declined to hear an appeal in March 2026, leaving the D.C. Circuit's holding standing, though that's a denial of certiorari rather than a ruling. But AI-assisted work with genuine human authorship remains registrable, and hundreds of such works have been. Your platform's terms assign whatever rights it holds and cannot create copyright the law denies, so "we own it, the terms say so" answers the wrong question. The real risk for marketers is usually being unable to stop others copying rather than being sued — which matters for brand assets and barely at all for blog posts. Jurisdictions genuinely differ, with UK and Indian provisions written decades before generative AI. Document what a person actually decided, at the time, for anything you'd need to defend.
Building brand assets you might one day need to defend?
We create original brand and design work with clear ownership, so the question never arises.
Explore Content Marketing →Frequently asked questions
Can you copyright AI-generated content?
In the United States, not where the work has no human author. The D.C. Circuit held in Thaler v. Perlmutter that eligible work must be authored in the first instance by a human being, and the Supreme Court declined to hear an appeal in March 2026, leaving that holding standing. Importantly, the court also clarified that this does not prohibit copyrighting work made with the assistance of AI — it requires only that the author be the human who created, operated or used the system rather than the machine itself.
Does a platform's terms of service give you ownership of AI outputs?
It can assign whatever rights the platform holds, but it cannot create copyright protection that the law does not grant. This distinction catches people out because the terms genuinely do say ownership transfers to the user. What that means in practice is that the platform is not claiming rights against you — not that a protectable copyright exists. If the output has no human author, there is no copyright for anyone to transfer, and a contract between two parties cannot manufacture one.
What part of AI-assisted work is actually protectable?
The human contribution. Registration practice in the United States has covered human-authored text combined with AI-generated images, substantial human modification of AI output, and the creative selection, coordination and arrangement of generated material. The Copyright Office has reported registering hundreds of works incorporating AI-generated material, with protection extending to the human author's contribution rather than to the generated elements themselves. The practical implication is to document what a person actually decided and did.
How do the rules differ between countries?
Considerably, and this matters for anyone publishing internationally. The United Kingdom has a long-standing provision attributing authorship of computer-generated works to the person who made the arrangements necessary for their creation. India's Copyright Act contains a comparable provision defining the author as the person who causes the work to be created, though how it applies to generative systems remains unsettled. Chinese courts have been more willing to find protection where creativity traces to human input, while the US position centres firmly on human authorship.
What is the real commercial risk for marketers using AI content?
Frequently the inability to stop others copying it, rather than being sued. If a piece of output has no human author, it may have no copyright owner, which means a competitor could reproduce it without infringing anything. For most marketing content that is a minor concern, but it matters considerably for assets intended to be distinctive and defensible — brand imagery, signature visual identity, or work you would expect to enforce against imitation. The separate question of training-data liability sits with model providers rather than users in most circumstances.
