Putting the “Art” in Artificial Intelligence: Does Copyright Law Protect AI-Generated Work?

The Constitution provides that authors and inventors should be able to control their works for an exclusive period of time, with the premise that giving this exclusive right promotes progress in both science and “the useful Arts.” Until now, this clause has served us well. Will it continue to do so in the age of Artificial Intelligence?

Discussions of AI and copyright tend to focus on its generative capabilities, in which computers are fed massive amounts of existing, human-created material. Computer algorithms learn from this material until they can create new versions on their own when prompted. Now, there’s an AI that creates “new Rembrandt” paintings—new images based on the Dutch master’s work—and ChatGPT can instantly generate text that mimics any source from Shakespeare to Jerry Seinfeld.

In March 2023, the Copyright Office declared that AI-created material cannot be copyrighted. In many ways, the complex new policy raises as many questions as it answers.

Some Copyright Basics

Over the years, Congress has passed various versions of the Copyright Act, which charges the U.S. Copyright Office with registering copyright applications. While the law dictates aspects of copyright, it also allows the Copyright Office to create additional requirements for copyright filings. These rules are allowable as long as they are consistent with the law.

One way of testing a rule’s consistency is in federal court. If an applicant has been denied a copyright, for instance, they can sue the Copyright Office, challenging that the Office has misapplied the law. If the court upholds the Office’s determination, revising its policies and practices takes an act of Congress.

Notably, there are longstanding limits on what can receive a federal copyright registration: You can’t copyright a title, name, or short phrase alone. You can’t copyright a design layout. You cannot copyright an idea—only how the idea is presented. You can’t copyright someone else’s creation unless you own the rights to their work.

You also can’t copyright a process. A recipe can’t be copyrighted if it only consists of a list of ingredients or the process of preparing the dish. But you can copyright the photographs or written description that accompany the recipe, as well as the story you tell about it. In the case of software, only the code can be protected, not what it does.

AI Creations Can’t Be Copyrighted When They Lack a Human Author

The Copyright Office’s policy says that the author of a work must be human to receive a copyright. Non-humans are ineligible. In its March policy statement, the U.S. Copyright Office clarified this position: AI-generated materials are not eligible for copyright because they lack a human author.

The courts have upheld the Copyright Office’s “humans only” policy. For example, in 2016, an appellate court famously concluded that a monkey could not hold the copyright to photographs he’d taken with someone’s camera. Other nations—such as Germany and Spain—also limit copyright protection to human registrants.

Thus, the Copyright Office recently refused to register a copyright for “A Recent Entrance to Paradise,” an image solely created by “The Creativity Machine,” an AI system owned by Dr. Stephen Thaler.

In his application, Thaler stated that the algorithm had created the image, and he was registering it on the algorithm’s behalf. The Copyright Office denied the application on the grounds that the image lacked a human author. A subsequent review board affirmed the Office’s decision.

Thaler is now suing the Copyright Office in federal court, arguing that the requirement for human authorship is not part of the Copyright Act that governs the department. New technologies are likely to continue to challenge the Office’s provision on human authorship—as they’ve done since its inception. In the 1800s, for instance, the courts grappled with whether the author of a photograph was the camera or the photographer.

It’s unlikely that Thaler will prevail in the present case, especially given that the Supreme Court has refused to hear it. This isn’t his first loss on behalf of AI “authors”: The U.S. Patent Office has twice denied Thaler’s earlier applications on the grounds that inventors, like authors, must be human.

Can’t Thaler simply obtain a copyright because he created the AI program? No, it turns out. As commentators have observed, this argument would likely fail since it’s akin to saying that Microsoft not only has a copyright on Microsoft Word, but it can also own the content of every document created with the program.

But What If There IS a Human Author Involved?

Thaler’s applications, which he files to test the laws on AI-generated material, deliberately exclude the possibility of a human author. But there’s a broad middle ground between entirely human-generated and entirely AI-generated work: collaborative endeavors in which a creator uses AI as an aide in the creative process.

Currently, the leading reference point is the Copyright Office’s decision in Zarya of the Dawn. In that case, the Office had initially granted Kristina Kashtanova a copyright for Zarya of the Dawn, a comic book. Kashtanova had written the text for the book herself but used AI to generate its images. The Copyright Office canceled her registration and reissued a registration limited to her text alone.

Kashtanova argued that she controlled the AI as a tool: the images in the book were the result of her prompts, selections, and editing. The Copyright Office concluded that such control was insufficient because, ultimately, AI generates products in an unpredictable way.

Thus, for the Copyright Office, it is more appropriate to think of the AI-created product as a work generated by a commissioned artist—what it calls a work-for-hire—rather than a work created by an artist wielding a tool. Since the commissioned artist wasn’t human, the applicant is ineligible for copyright.

In the March policy, the Copyright Office clarified its approach, stating that if an author relies on AI for anything beyond a de minimus use when creating a registered work, the author must clearly identify that AI-generated work in the application. If the larger work is copyrighted, the AI-generated contributions will be excluded from the registration.

This stance makes the Office’s position on AI analogous to its position on recipes and other mixed work. Original contributions by humans are protected; the AI-generated material in the work is not. The question of what constitutes de minimus use will be determined on a case-by-case basis—surely a catalyst for future litigation.

Authors and artists with pending registration applications that don’t disclose AI’s involvement in their work must revise their applications or risk denial.

Do AI Creations Infringe on Artists’ Copyrights?

The more complex legal question for authors and artists using AI is whether they can be sued for copyright infringement.

By definition, generative AI is based on original work by humans—including material under copyright. In a number of pending cases, copyright holders are suing creators for unauthorized use of their work in the process of generating their AI products. Some of these cases focus on the use of Open Source code. The very point of Open Source material is that it can be used by anyone for free—a practice intended to innovation and creativity. But Open Source creators can require users to credit their work, and the plaintiffs allege that the AI products don’t sufficiently credit their contributions.

Similarly, there are also cases of artists suing AI users for copyright infringement when they deploy the technology to mimic their work. For example, Universal Music Group is suing platforms over an AI-created song, “Heart on My Sleeve,” which was created to emulate the work of Drake and The Weeknd.

These cases may be more difficult to win because infringement can be difficult to prove. For example, parodies are usually protected as new creations, as is anything that substantially transforms the original work. This is the reason fan fiction—new stories based on existing characters—is so prevalent. Fan fiction openly draws on copyrighted material, but if there’s enough new content, owners of the original source have a difficult time prevailing in court. Going forward, it could become difficult to say why human-created fan fiction is allowable but computer-generated fan fiction is not.

Copyright is not the only way artists and authors can protect their material. Depending on the facts involved, other statutes could offer more direct solutions. For example, under the Lanham Act, which protects against false designation of a product’s origin, a user may have to affirmatively disclose the use AI.

Legal Solutions to AI-Use Questions

As AI technology continues to develop, so will the laws governing ownership and copyright. If you are an artist dealing with AI-use questions—whether you are using it to create work or you believe your work has been infringed—this is one of those times when you should speak to a real, live human. To do that, please contact Carrie Ward at (856) 354-7700 or at cward@earpcohn.com.