top of page

Who Owns AI Generated Art?

  • Writer: Nikita Silaech
    Nikita Silaech
  • 3 days ago
  • 2 min read
ree

Stephen Thaler created a painting using an AI system he built and called it "A Recent Entrance to Paradise," thinking it was good enough to copyright. The US Copyright Office said no, arguing that copyright requires human authorship and a machine cannot be an author (CNBC, 2025).


Thaler appealed, arguing that he created the AI and guided it, so he was the author just like a photographer is the author of a photograph even though the camera does the work.


The court disagreed. The panel ruled that at least some degree of human authorship is essential for copyright eligibility. The person might have created the AI, but they did not create the specific image, the machine did.


So, if Thaler does not own the copyright, who does? The AI cannot own it, and nobody else created the image. Thaler called this orphaned intellectual property, a work that exists but belongs to nobody. He asked the Supreme Court to reconsider.


Meanwhile, three artists sued Midjourney, Stability AI, and DeviantArt in 2023, saying these companies trained their models on billions of copyrighted images without permission. The argument was that these companies scraped their work and used it to generate outputs that imitated their style. That is copyright infringement (The Verge, 2023).


The companies argued that it was fair use, saying training on existing work is similar to how photographers and digital artists have always learned by studying others' work.


A judge allowed the case to move to discovery in August 2024, finding the copyright infringement claim plausible (JIPEL, 2024). But here is where it gets confusing. Even if the court finds that training on existing art is infringement, what is the remedy? The AI companies argue that they do not store the images but rather extract mathematical representations of patterns. If they are not storing the images, can they have copied them in the traditional sense?


This is a question courts have never had to answer before. Then there is the question of the output itself. Suppose someone generates an image using Midjourney and styles it after Monet. Is that copyright infringement? Copyright does not protect style, only specific works. The question becomes whether the output contains substantial similarity to a specific original work, and an AI system can generate a painting that looks Monet-like without being a copy of any specific Monet.


The European Union passed the AI Act in 2024, requiring AI companies to get permission from copyright holders if they use copyrighted content for training. The United States didn’t have an equivalent law at the time. The Copyright Office had said that human-generated elements can be copyrighted separately from AI-generated elements, so a comic book with human-written text and AI-generated illustrations can have partial copyright protection.


But what about a painting where a human wrote a prompt and an AI rendered the image? How much human creativity is required for copyright to exist? Nobody knows.


Courts will eventually have to decide, but every decision will likely create more questions than it answers. Copyright law was built around human creation, and AI is forcing that system to confront what it means to be an author at all.

Comments


bottom of page