Who owns AI-made work in 2026 - a producer's view of the copyright question

The rule after March 2026

For a while, “who owns AI output” was a question you could answer with a shrug and a hope. That window closed in March 2026, when the US Supreme Court declined to hear Thaler v. Perlmutter, leaving intact the position the Copyright Office has held all along: copyright requires a human author. A machine cannot be one.

The practical consequence lands hard on anyone shipping brand assets. Output that is purely machine-generated - a prompt in, a file out, no meaningful human authorship - has no copyright owner. Which means, in effect, it is public domain. Anyone can take your one-prompt hero image and run it in their own campaign, and copyright law will not help you stop them.

Meanwhile the litigation environment stays loud: legal trackers count over seventy active AI copyright lawsuits in the US alone, most about training data. The ownership rule, though, is now stable enough to build a process on.

Generated versus assisted - where the line sits

The law distinguishes two things the market keeps blurring.

AI-generated: the machine did the creative work. Prompts alone do not count as authorship - the Copyright Office has been consistent on that since the Zarya of the Dawn registration case, where images made from prompts were refused protection even though the human wrote every prompt.

AI-assisted: a human made real creative decisions and used the model as a tool. Selection and arrangement, substantial editing, compositing, original elements layered in, direction of a larger whole - that human contribution is protectable, in the same way a photographer owns a photo despite the camera doing the physics.

The line is fuzzy at the edges and will be litigated for years. But its direction is friendly to exactly one kind of practice: the kind with lots of documented human decisions between the first prompt and the final file.

Platform license is not ownership

A confusion worth killing: the license a platform gives you and the copyright status of the output are two separate questions. A paid plan on an image or music platform can grant you full commercial rights to use what you make - and that same output can still have no copyright owner because no human authored it. You may sell it. You may not stop others from using it.

The platforms differ on the license half too - some gate commercial use behind paid tiers, one large image platform requires a higher plan above a million dollars of annual revenue, and terms shift. Read the tier, not the homepage. And note the registration trap reported in the US: registering AI-assisted work without disclosing the AI involvement can void the registration. Honesty is not just ethics here; it is validity.

What a rights-aware pipeline looks like

This is where a legal question becomes a production question, which is our home turf. If human authorship is what creates ownership, then a studio’s pipeline is an ownership-generating machine - provided the human authorship is real and provable. What that means for us in practice:

Every project keeps its decision trail - reference decks, style frames, edit versions, retouch layers, the record of what was chosen and rejected. Human craft is layered into every deliverable: composition, edit, grade, sound, finishing. And the delivery package can state plainly which parts carry human authorship. When a client’s legal team asks “what do we own”, a disciplined studio can answer with evidence instead of adjectives.

We don’t generate. We produce. It began as a craft statement. In 2026 it is also the ownership strategy.

The view beyond the US

Short version for anyone shipping internationally: the EU matches the human-authorship principle and adds training-data transparency duties under the AI Act as it phases in, with proposals this spring for a training-data register and opt-out rights. The UK is the odd one out, with a statute granting limited protection for computer-generated works. Canada and Australia hold the no-human-no-copyright line pending reviews; Japan leans most permissive, case by case. If a campaign crosses borders, assume the strictest applicable rule and let the pipeline documentation do the work everywhere.