Copyright use by AI: pioneering cases that may define tomorrow’s solutions

What just a few years ago seemed to be a matter of science fiction, the creation of images and other content by artificial intelligence (AI) is now a reality. As AI-generated images disseminate across the internet, many concerns arise in the real world, especially with regard to copyright protection. In this scenario, while we are still trying to understand how to regulate AI, a battle between authors and copyright owners against AI generative technologies has been set¹.

Two pioneering legal cases recently filed in the US and the UK contesting the use of copyright content to “train” these technologies illustrate well the conflicts arising between copyright and AI and the complexities surrounding it. Their outcome will be decisive to indicate how the law will deal with the challenges posed by these revolutionary technologies.

‍

Andersen et al v. Stability AI Ltd. Et al.²

In January 2023, the artists Sarah Andersen³, Kelly McKernan⁴, and Karla Ortiz⁵ filed a class-action lawsuit before the California Federal Court against Stability AI, Midjourney, and DevianArt⁶, accusing them of committing mass copyright infringement by “scraping the internet” to copy and store billions of copyrighted images without obtaining consent or licenses from artists and then using the copied images as input for training their AI platforms, without the artists knowledge or consent.

According to the artists:

• The defendants are “using copies of the training images interconnected with their AI image products to general digital images and other output that are derived exclusively from the training images, and that adds nothing new”;
• The output images are exclusively derived from copies of copyrighted images and, therefore, are necessarily derivative works. That is why they say that these AI image-generative technologies are, “in short, a 21st-century collage tool”.

Thus, the artists seek to end what they consider as “an enormous infringement of their rights” before their professions are substituted“ by a computer program powered entirely by their hard work."

In response, the defendants alleged that, in general, their tool gets the "idea” or the “style” of previous works to do their own works. Thus, as the idea cannot be protected by copyright, there wouldn’t be any infringement, arguing that:

• The artists’ allegations represent a misunderstanding of how their technology works and the law. The platform Stability Diffusion would “enable users to create entirely new and unique images utilizing simple word prompts”, not infringing copyrighted works;
• Their tool would be similar to a human artist: they would learn from a picture already made and do one on their own. “To the extent that AIs are learning like people, it's sort of the same thing and if the images come out differently, then it seems like it's fine," he said.”
• The artists and their lawyers would be trying to establish copyright over style, “something that has never before been legally protected”.

Another issue pointed out by the AI companies’ lawyers is that most of the works alleged to be infringed do not hold a copyright registration before the US Copyright Office, which would be a major obstacle for their claims⁷.

‍

A positive response to the AI companies

During the hearing held last July 19th, U.S. District Judge William H. Orrick ruled that the plaintiffs failed to present clearly the facts and how each defendant could be held liable for copyright infringement. He expressed that he didn’t think that the claim would be plausible for the moment, stating that there’s no substantial similarity between the Artists' images and the images generated by the AI tools.

Nevertheless, the judge allowed the plaintiffs to amend their complaint. Hence, the legal action is still ongoing.

‍

Getty Images vs. Stability AI⁸

In February, Getty Images⁹ filed a lawsuit against Stability AI in the U.S. District Court of Delaware claiming that their AI art tool copied and processed 12 million images, associated text, and metadata to train its model, without requesting any license. In May, a second lawsuit was filed in London's High Court of Justice of England and Wales¹⁰, to prevent Stability from selling its AI-image generator tool in the UK.

According to the Getty Image, Stability would have committed:

• Copyright infringement due to the alleged "reproduction" of the copyrighted images and the distribution of the synthetic images to the public without Getty Image’s consent;
• Provision of false copyright management information, removal or alteration of copyright management information;
• Database rights infringement, for training and "re-utilizing" the images and respective captions;
• Trademark infringement and dilution, as generated images occasionally incorporate "substantial parts" of copyrighted images and, sometimes, even feature Getty Images watermarks;
• Unfair competition and deceptive trade practices under the Delaware State laws;

Up to now, Stability limited its argument to a procedural matter: according to its motion, the District Court of Delaware lacks jurisdiction over the dispute, as Stability’s principal place of business is London. Thus, the AI company requested the legal action be dismissed or transferred to the Northern District of California where the artists’ class action is pending.

However, Stability’s possible line of defense is the fair use doctrine (art. 17, §107 US Copyright Act), which permits limited use of copyrighted material without having to first acquire permission from the copyright holder. As it requires a multi-factor and case-by-case analysis¹¹ to be applied, it is still uncertain how the courts will apply this doctrine in the context of AI and deep learning.

In fact, the solution to both of these cases depends mostly on the interpretation of the AI-generative system’s mode of function, which according to Jane Ginsburg¹² may have two different versions:

• “The AI system scoops up all these images and the system then 'learns' what these images look like so that it can make its own images”; or
• “The system is not only copying, but it's also pasting portions of the copied material, creating collages of the stored images”.

‍

Challenges for a very near future

Whichever will be the outcome of these cases, they will certainly represent major precedents not only for the solution of other legal cases but also may indicate how the lawmakers will regulate the copyright use by AI. As our current copyright and, most generally, intellectual property legislations are not capable to cover the uncertainties brought by AI, its regulation is urgently expected.

Do not hesitate to contact our IP/IT team in case of any doubt.