Can You Own the Copyright to AI-Generated Content?

Can You Own the Copyright to AI-Generated Content?

Generative AI can turn a short prompt into an image, a video, or a piece of writing. Yet the ease of producing content does not make the rights surrounding it equally straightforward.

If you write the prompt, do you own the resulting image? If an AI-generated image is not protected by copyright, can you freely use it in an advertisement or a commercial product?

Kyungsook Kim’s 2025 paper on generative AI and copyright examines these questions through two distinct perspectives. The first concerns copyright infringement during the training, generation, and use of AI systems. The second concerns copyrightability: whether the resulting content itself qualifies for protection.

These questions are related, but they require different assessments. Understanding that distinction is the starting point for this review.

1. Infringing Existing Rights and Establishing Your Own Rights

Copyright infringement begins with the rights attached to existing works. Relevant questions include whether training materials were lawfully used and whether an output is substantially similar to protected expression in an existing work.

Copyrightability begins with the newly produced content. It asks whether human creativity is present and, if so, which elements may receive protection.

An AI-generated output may therefore lack copyright protection while still potentially infringing someone else’s copyright. Conversely, a person may hold copyright in their creative additions without having permission to use third-party expression incorporated into the same work.

“Can I protect this?” and “Can I use this?” must be considered separately.

2. Training and Output Use Raise Different Issues

The paper examines copyright issues at different stages rather than treating generative AI as a single act.

During training, a central issue is the reproduction of existing works when collecting and processing data. The analysis may involve permission from rights holders and exceptions or limitations such as fair use and text-and-data-mining provisions. Because these rules differ across jurisdictions, AI training cannot simply be classified as universally lawful or unlawful.

At the generation and use stages, attention shifts to the actual output. The inclusion of a work in training data does not, by itself, establish that every resulting output infringes that work. The relationship between a particular output and an existing work must be examined.

Platform responsibility is another distinct issue. Direct involvement in training may raise different questions from inducing or facilitating infringement by users.

The statement “it was made with AI” is therefore insufficient to determine the responsibility of a user or service provider.

3. Human Creativity Remains Central to Copyright Protection

When assessing copyrightability, the paper emphasizes human creative contribution.

The presence of AI-generated elements does not necessarily exclude an entire work from protection. If a person uses AI as a tool and contributes creative expression, that contribution may qualify for protection.

Time and effort alone, however, do not settle the issue. Spending hours writing prompts or repeatedly generating outputs does not automatically establish copyright in the final result. The question is how human creative choices and expression are embodied in that result.

Drawing on the U.S. Copyright Office’s approach, the paper discusses three main forms of human contribution:

• Creatively selecting, coordinating, or arranging generated material.

• Adding original expression through modifications or further creative work.

• Incorporating human-authored expression that remains perceptible in the output.

4. Protection May Cover Only Part of a Work

The first form of contribution involves creativity in selecting and arranging generated material.

The paper discusses Zarya of the Dawn, a comic incorporating Midjourney-generated images. Copyright protection covered the human-authored text and the selection and arrangement of text and images, while the individual AI-generated images were excluded.

The Korean work AI Surobuin provides a related example. Its registration as a compilation recognized creativity in the selection and arrangement of AI-generated material. This should not be interpreted as recognition of copyright in every AI-generated element within the video.

The second form involves creative modifications. When a person adds original expression to an AI output, those contributions may qualify for protection. Routine technical adjustments and changes that introduce creative expression should nevertheless be distinguished.

The third form involves human-authored expression that remains identifiable after AI processing. In the Rose Enigma example discussed in the paper, the key issue was whether expression from a person’s original sketch remained recognizable in the resulting image. Protection focused on identifiable human authorship, rather than all the details newly produced by AI.

These cases show why copyright protection is not always an all-or-nothing question. Different elements within the same work may receive different treatment.

5. Does a Better Prompt Mean a Stronger Copyright Claim?

Prompt writing is an important part of using generative AI. But describing a desired result is not always equivalent to creating its specific expression.

Under the U.S. approach examined in the paper, a key question is whether prompts provide sufficient human control over the expressive elements of the output. Even detailed instructions may leave the system to determine the specific composition and visual details. In such circumstances, the instructions alone may be insufficient to establish human authorship of the entire result.

The paper also shows that outcomes can differ across jurisdictions and cases. Broad claims—such as “prompts never matter” or “a certain number of revisions creates copyright”—should therefore be avoided.

The central issue is the human creative expression reflected in the final work, rather than prompt length or the number of generation attempts alone.

6. What This Means for Content Production

For content creators, the paper suggests two separate areas of attention.

First, consider whether the output incorporates protected expression belonging to someone else. Personally generating an image or paying for access to an AI service does not, by itself, resolve that question.

Second, be able to explain your own creative contribution. Keeping original drafts and sketches, documenting selection and arrangement decisions, and retaining versions before and after editing may help demonstrate what a person created. This is a practical implication of the case analysis, rather than a guarantee that documentation will establish copyright.

Alongside identifying the AI tool used, creators increasingly need to explain what expression they themselves contributed.

7. The Paper’s Contribution and Limits

The paper’s contribution lies in separating infringement from copyrightability and using legal frameworks and cases from different jurisdictions to explain how these questions can be assessed.

It is a legal study based on materials examined for its 2025 publication. It does not experimentally measure human contribution or establish a numerical threshold based on prompt length or revision count. Differences in national law and individual circumstances also mean that a foreign registration decision cannot automatically determine the treatment of every work in Korea.

The paper encourages a closer examination of human creativity and specific acts. Rather than asking only how much AI produced, we should ask what expression a person created and how that expression remains present in the result.

Obtaining an output is only the beginning of the copyright analysis. The creative process, the final expression, and the relationship to existing works all matter.