Who owns what it makes

The question from the other side

The previous lesson was about what goes in. This one is about what comes out: when a model generates an image, a paragraph or a song, who owns it? The answer decides whether the logo you generated for your business is yours, whether a competitor can copy it, and whether the novel you drafted with a model can be registered. It differs by country, and the differences are instructive.

The United States: a human must have made it

American copyright requires human authorship, and the copyright office has said so repeatedly since 2023. In the case that set the pattern, a comic book with a human-written story and model-generated images was registered for the text and the arrangement, and refused for the images. A man who tried to register an image he said his system had created autonomously was refused, and in 2025 an appeals court agreed: no human author, no copyright.

The office’s 2025 guidance drew the practical line. A prompt alone, however detailed, does not make you the author of the output, because the model decides the expressive choices. Selecting, arranging and substantially editing generated material can produce a protectable work, in the parts you made. So a generated logo, unedited, may be unprotectable, which means anyone may use it, and a trademark registration becomes the way to hold it.

The United Kingdom and India: an older answer

Both countries wrote a clause for computer-generated works long before any of this. The British act of 1988 says that for a work generated by computer with no human author, the author is “the person by whom the arrangements necessary for the creation of the work are undertaken”, and protection lasts fifty years. India’s act, amended in 1994, says the author of a computer-generated work is “the person who causes the work to be created”.

Those clauses were written for the software of the time and nobody is certain how they apply to a prompt. Who made the arrangements: the lab that trained the model, or the person who typed? In India, in 2020, the copyright office registered a painting with an AI system named as co-author alongside a person, then issued a notice proposing to withdraw it; the position has not been clarified since. Britain’s government reviewed its clause in 2021 and kept it, then, after a further consultation, proposed in March 2026 to remove it; no law has changed yet.

The European Union: the author’s own creation

European law requires a work to be the author’s own intellectual creation, which courts have read as requiring human choices. There is no computer-generated-work clause. The prevailing view is that a purely generated work is not protected, and that a human who makes substantial creative choices in prompting, selecting and editing may own the result: the same shape as the American line, reached from different law.

China: protection, with human input

Chinese courts went the other way. In late 2023 a Beijing court held that an image generated from a detailed prompt with many rounds of parameter adjustment reflected the person’s intellectual input and was protected, and ordered a copier to pay damages. Later decisions have followed it. China’s answer, so far: substantial human input in the prompting makes the person the author.

What this means when you generate something

Four practical consequences, none of which is legal advice.

Your prompt is not a copyright. In most places, typing a description does not make you the author of what appears.

Unedited output may belong to no one. Which means it protects no one: a competitor may copy your generated logo without infringing, in the United States and probably Europe.

The terms of service are a contract, not a copyright. The major labs assign you whatever rights exist in the output. If none exist, the assignment is of nothing.

Disclosure is increasingly required. Journals, competitions, publishers, some funders and some courts ask whether material was generated, and the previous module’s copyright policy from the EU sits alongside a growing set of institutional rules.

Why this is unsettled and will stay so for a while

The law in every jurisdiction was written for a world in which making a work required a person. A model breaks the assumption in a way the statutes did not anticipate, and the responses (no protection, protection for the person who arranged it, protection for sufficient human input) are three different values about what copyright is for. They will converge slowly if at all, and a business that spans borders should assume the answer differs on each side of them.