Can You Copyright What an AI Made? The Human-Authorship Problem

You spend an hour crafting the perfect prompt. You refine it, tweak the wording, run it a dozen times, and finally the AI produces an image that is exactly what you imagined. It is striking, original-looking, entirely new — and, as far as copyright is concerned, it may belong to no one. This is the strange and increasingly urgent problem at the heart of AI-generated work: the tool can produce something that looks completely original, yet the law may refuse to grant it copyright protection at all. The reason comes down to a single, deceptively simple requirement that predates artificial intelligence by more than a century — human authorship — and understanding it is essential for anyone creating, sharing or relying on AI-generated content.

Copyright has always required a human

To see why this is such a problem, you have to understand a principle so foundational that until recently it barely needed stating: copyright protects the creative work of human authors. It exists to reward and encourage human creativity, and its protections have historically been reserved for works that originate from a human mind. This was never controversial, because for the entire history of copyright there was no other kind of author to worry about. A painting had a painter, a novel a novelist, a photograph a photographer. Human authorship was assumed because there was no alternative.

Artificial intelligence broke that assumption, and in doing so exposed a requirement that had always been there but never tested. When an image or a piece of text is produced by an AI system, the old, quiet premise that behind every work stands a human creator suddenly cannot be taken for granted. The copyright authorities responsible for these questions have had to confront, directly, whether something generated by a machine can be an authored work at all — and their answer has been clarifying, consistent, and, for many people hoping to own their AI output, disappointing.

The rule, stated plainly

The guidance that has emerged is unambiguous on the core point: a work that is entirely generated by artificial intelligence, without meaningful human creative input, is not eligible for copyright protection. If you give an AI a description and it produces the work, with the machine doing the actual creating, the output has no human author in the required sense, and therefore no copyright. It falls, in effect, outside the system of protection entirely — not because anyone decided AI work is unworthy, but because there is no human author to whom the law can attach the rights.

The most important and widely misunderstood consequence concerns prompts. Many people assume that because they wrote the prompt — sometimes a long, careful, elaborately engineered prompt — they authored the result. The authorities have been clear that this is not so: providing prompts to an AI, no matter how detailed or skilfully written, does not by itself make you the author of what the AI produces. The reasoning is that when you write a prompt, you are describing what you want and letting the machine make the countless creative decisions that actually produce the work — the specific forms, colours, arrangements, words. You expressed a wish; the AI did the creating. And copyright attaches to the creating, not the wishing. That distinction is the entire crux of the matter, and it dashes the most common assumption people bring to AI tools.

Where the human comes back in

This does not mean AI has to be a copyright dead end, and the more hopeful part of the guidance is just as important. Works that combine human and AI-generated elements can qualify for protection — but only for the parts a human genuinely authored. If a person uses AI as one tool among others and exercises significant creative control over the final result — meaningfully selecting, arranging, editing, altering and shaping the output so that their own original expression is present in the finished work — then that human-contributed expression can be protected, even though the raw AI-generated material on its own cannot.

The useful analogy the authorities themselves reach for is the camera. A camera is a machine that does an enormous amount of the technical work of making an image, yet photographs are copyrightable because the human photographer makes the creative choices — what to point at, how to frame, when to capture, how to treat the result. AI can be used similarly: not as a genie that produces the finished work from a wish, but as an instrument in the hands of a human who exercises real creative control over the outcome. The line, then, is not "did a machine touch this?" but "did a human contribute enough original creative expression to be an author?" Where the answer is yes, protection follows for that human contribution; where the human merely prompted and accepted, it does not. This is a distinction of substance, and it connects to the broader questions of ownership and reuse we examined in Creative Commons licenses explained: you can only license, or protect, what you actually own.

The disclosure duty and the practical reality

Because the line runs between human and machine contribution, the systems that register creative works have added a practical requirement that flows directly from the principle: honesty about what the machine made. Anyone seeking protection for a work involving AI is expected to disclose the AI-generated material and to identify the human contributions they are actually claiming, rather than presenting a machine-made work as though it were wholly their own. You cannot quietly launder AI output into a fully protected human work by staying silent about how it was made; the claim has to be limited to the genuinely human parts, and the AI-generated portions have to be acknowledged as outside the claim.

In practice, this has produced a workable but demanding middle ground. A great many works blending human and AI effort have been successfully registered — not as fully protected AI creations, which do not exist, but as works whose human-authored elements are protected while the purely AI-generated portions are not. The effect is that the more a human genuinely shapes, selects, arranges and transforms AI output into something bearing their own creative expression, the more of the result they can protect; the more they simply accept what the machine produced, the less. This is not a loophole to be gamed but a coherent application of the century-old principle: protection scales with genuine human authorship, and the disclosure duty exists to keep that honest.

Why this matters for open culture and everyone else

The human-authorship problem is not an abstract legal curiosity; it has real consequences for how AI-generated content can be used, shared and relied upon. A vast and growing amount of the imagery and text circulating online is now AI-generated, and much of it, under the principle described here, has no clear copyright owner at all. That has practical implications in both directions. For creators, it means that output you generated and feel is "yours" may not be protectable, and that others may be free to use it in ways you cannot prevent. For anyone reusing such material, it means the usual assumption that a striking, professional-looking work must belong to someone — and must therefore be licensed — may simply not hold.

This is where the question touches open culture most directly. A world in which enormous quantities of content have no human author, and therefore no copyright, is genuinely new, and the frameworks built to manage the sharing of human-made work were never designed for it. It unsettles the basic mechanics of licensing, attribution and permission, because those mechanics assume an owner exists. The honest position for now is caution rather than certainty: treat AI-generated material as legally uncertain, understand that a purely machine-made work likely has no copyright owner to license or protect it, and recognise that the amount of genuine human creative control involved is what determines whether anyone owns it at all. The same care applies to the boundaries of reuse we explored in what fair use actually means and where it ends.

The line that will define the next era

The deepest lesson of the human-authorship problem is that AI has forced copyright back to its first principle and made everyone look at it plainly. For a hundred years the requirement that a work have a human author sat unexamined, because there was never any other kind of author. Now there is, and the law's answer has been to hold the line: protection is for human creativity, the machine's output alone earns none, and the more genuine human authorship a work contains, the more of it can be owned. It is a principle old enough to feel almost quaint and suddenly more consequential than it has ever been.

For creators, the practical takeaway is clear and worth internalising before you rely on anything you have generated. If you want to own what you make with AI, you cannot simply prompt and accept; you have to bring real creative control — selecting, shaping, arranging, transforming — so that your own original expression genuinely lives in the result. Treat AI as a tool you wield, not an author you commission, and be honest about which parts the machine made. Do that, and the human-authored expression you contribute can be protected. Skip it, and you may have made something striking that belongs, in the eyes of the law, to no one at all. In an age flooding with machine-made content, that century-old requirement of a human author is quietly becoming one of the most important lines in all of creative work.