Various generative AI tools can create text, images, songs, videos and other content, and regulators, courts and vendors are coming to terms with what this technology means for copyright law.
If an organization wants to incorporate AI-generated content into its marketing or content strategies, it must consider two broad questions to avoid copyright infringement:
- Can AI-generated content be copyrighted?
- What is considered copyright infringement for AI-generated content?
Copyright law, as it currently stands, was written at a time when humans were directly involved in fair use practices, like citing sources and creating derivative works, which is copyrighted work that comes from other copyrighted work. New AI models can scan copyright-protected content at scale to distill an image's style, a novel's plot or a program's logic. Once trained on protected content, these AI models can generate new content different enough from the original that some might consider it fair use.
Yet, users of popular generative AI platforms can't determine how these services were trained.
"The problem with AI-generated content is that users don't know exactly where the AI is sourcing things from and which parts of the content it creates are from scratch or just pulled from another piece of copyrighted content, or even another person's AI-generated artwork on the platform," said Nizel Adams, CEO and principal engineer at Nizel Corporation, an AI consultancy.
This article is part of
Can AI-generated content be copyrighted?
If users of generative AI platforms don't know what content the tools were trained on, then they may hesitate to adopt these platforms in case their AI-generated content is already copyrighted and doesn't fall within fair use.
The U.S. Copyright Office has recently provided some guidance on this topic, but the overarching answer is: AI-generated content can sometimes be copyrighted, according to David Siegel, partner at Grellas Shah LLP. Siegel said he expects more guidance as the courts tackle the issue.
Thus far, the Copyright Office, in line with existing case law, has explained that, for a work to be afforded copyright protection in the U.S., it must have a human author. Yet, Siegel said he is not sure what that means in the world of AI.
"If the only human involvement is the input of a chat prompt into ChatGPT, for example, one cannot obtain copyright protection for the raw result of that prompt," he said.
On the other hand, if a user inputs a prompt into an AI tool, gets a response and then modifies the result in creative ways, that can potentially result in content afforded copyright protection. However, only human-authored parts of the work can be copyrighted.
In other words, AI can be a tool authors use to generate materials and create copyrighted works. "But that is a far cry from what most people think about when considering whether AI-generated content can be copyrighted, which is traditionally focused on copyrighting raw outputs," Siegel said.
What is considered copyright infringement for AI-generated content?
Users may wonder if AI-generated content trained on protected intellectual property is considered copyright infringement. This question has a murkier answer and is the subject of numerous lawsuits on images, songs and books. For example, in early 2023, TikTok, Spotify and YouTube removed an AI-generated song that mimicked the voices of rapper Drake and R&B artist The Weeknd. However, the implications for AI-generated content less similar than this example are unclear.
A copyright infringement inquiry begins with the long-established test of access and substantial similarity, according to William Scott Goldman, managing attorney and founder at Goldman Law Group. Overall, this means the case would have to prove that the AI or a human read the content and that it's similar enough to convince a jury it was copied.
"Although there is no established case law for generative AI just yet, I believe without clear-cut proof of access, such infringement claims will fail unless the copying in question is deemed identical to the original," Goldman said.
However, Goldman also said he believes copyright owners and plaintiffs could assert unauthorized use, especially if this use is not considered de minimis -- too small to be considered meaningful -- and the resulting work is substantially similar to the original.
Once both issues have been demonstrated, the case would turn on a fair use defense. Generative AI could be considered a derivative work under existing copyright law if it contains sufficient original authorship, Goldman said.
Courts now grapple with whether AI-generated content sufficiently differs enough from the originals under existing fair use precedents.
Lawsuits over AI-generated content
Various lawsuits regarding AI-generated content and generative AI tools have started to make their way through the courts -- both related and unrelated to copyright.
In November 2022, programmers filed a class-action lawsuit against GitHub, Microsoft and OpenAI focusing on breach of contract and privacy claims. In January 2023, the same law firm also filed a class-action lawsuit related to AI-generated image services, such as Stability AI's Stable Diffusion, Midjourney and DreamUp, which raises copyright infringement issues.
A few days later, Getty Images also filed a lawsuit relating to Stable Diffusion, arguing the service had "copied more than 12 million photographs from Getty Images' collection, along with the associated captions and metadata, without permission from or compensation to Getty Images," according to the lawsuit. In July 2023, Sarah Silverman and other authors sued OpenAI and Meta, claiming the generative AI training process infringed on the copyright protection of their works.
These lawsuits differ in important ways, according to Siegel. Stability AI allegedly uses images from the web to train its models. As a result, Getty's customers arguably have less need to license more images from Getty. This gets at the heart of copyright law, which is to incentivize people to develop creative works. Photographers and artists may be less willing to spend time and resources developing photos and images if those are used to train AI to replace them.
"If you are a photographer, would you be willing to spend your time and resources creating photos if those photos were going to be used to train an AI model, without compensation or permission, and potential licensees of your images could simply go to the AI model instead? Doubtful," Siegel said.
The Silverman case against OpenAI and Meta centers around the ability to provide summaries of books without permission to create derivative works from the authors. Siegel said this case differs from the Getty Images one because its use is similar to CliffsNotes, which is considered fine as people can still buy the book to get the full story.
How will AI change copyright laws?
The short-term future of AI-generated content copyright will likely aim to clarify existing concepts, like fair use and authorship for the AI age. Most AI companies rely on fair use to justify how they train their models, but it presents a gray area, Siegel said.
"To the extent the U.S. wants to foster the development of AI businesses, the laws around the use of copyrighted works in training AI models need to be sufficiently clear that even an early-stage startup can predictably determine whether their business model will run afoul of copyright laws. We are not even close to that point," Siegel said.
Overall, AI is changing copyright law. It is causing the legal system to define what constitutes authorship and how to protect human-generated content even if it contains AI-generated content, said Robert Scott, managing partner and Scott & Scott LLP.
In the U.S., the Copyright Office guidance states that works containing AI-generated content are not copyrightable without evidence that a human author contributed creatively. New laws can help clarify the level of human contribution needed to protect works containing AI-generated content.