Update as of 2/8/23: After beginning legal proceedings against Stability AI, Stability Diffusion’s creator, Getty Images is claiming in court documents that Stable Diffusion has stolen over 12 million of its copyrighted photos and is seeking up to $150,000 per infringement (about $1.8 trillion).

As generative art algorithms expand in popularity, artists and photographers have voiced concerns over how artificial intelligence programs are designed. Is it legal (or ethical) to train a software program by using millions of copyrighted images? The question will soon be debated in legal proceedings in both London and the U.S. On January 13, a trio of artists filed a complaint in California’s Northern District Court against Stability AI LTD., Stability AI Inc., Midjourney, Inc., and DeviantArt, Inc. over the use of copyrighted images in the training datasets. Just a few days later, stock image powerhouse Getty began legal proceedings in London against Stability AI. Both proceedings claim that the respective AI companies infringed on artists’ copyright when training the software.

In California, three artists—Sarah Anderson, Kelly McKernan, and Karla Ortiz—filed suit against the creators of Stable Diffusion and Midjourney, as well as DeviantArt for its new AI generator DreamUp. The complaint lists numerous questions for the courts to decide, including whether companies violated copyright law when downloading and storing copies of works and whether copyright was violated when using the defendants images “to create Fakes.” The complaint also names the Right to Publicity, questioning if the AI software’s ability to create images in the style of a named artist violates that right. The complaint also lists unfair competition laws.

[Read: Photographer Uses AI to Bring Dead Celebrities to Life]

“Hav­ing copied the five bil­lion images—with­out the con­sent of the orig­i­nal artists—Sta­ble Dif­fu­sion relies on a math­e­mat­i­cal process called dif­fu­sion to store com­pressed copies of these train­ing images, which in turn are recom­bined to derive other images,” wrote Matthew Butterick, counsel for the Plaintiffs in the California District Court case. “It is, in short, a 21st-cen­tury col­lage tool.”

This week, Getty also announced a lawsuit against Stability AI, the company behind Stable Diffusion, an AI-based tool that creates images based on text prompts. “It is Getty Images’ position that Stability AI unlawfully copied and processed millions of images protected by copyright and the associated metadata owned or represented by Getty Images absent a license to benefit Stability AI’s commercial interests and to the detriment of the content creators,” Getty wrote in a statement.

[Read: New AI Apps Raise Questions About Copyright and More]

Some users have found that the AI program reproduced the Getty watermark on generated images, suggesting a large number of photos with that watermark were used to train the software. In its statement, Getty also noted that it offers a specific license for training AI software. “Stability AI did not seek any such license from Getty Images and instead, we believe, chose to ignore viable licensing options and long‑standing legal protections in pursuit of their stand‑alone commercial interests,” the company wrote.

Since Stable Diffusion’s launch and subsequent popularity last year, artists have questioned the legality of using copyrighted works to train software without the owners permission or knowledge. But, as a new technology, the law has little precedence to help answer the questions that AI-generated art poises. The court cases filed this month could change that, allowing the courts to address those questions.

The companies behind generative AI have previously argued that the use of crawling the web for training images falls under Fair Use in the U.S. The way generative AI works is also often misunderstood. For example, generative AI doesn’t store forms of the actual training images used, but code related to the patterns in those images.

The legal proceedings initiated by Getty and the three artists aren’t the first to question how
artificial intelligence programs use existing data. Last fall, Micosoft’s GitHub CoPilot was part of a class action lawsuit claiming that the AI-based coding assistant contained chunks of code from existing copyrighted programs. The lawsuit filed in federal court claims that CoPilot contains the code of 11 open-source programs but fails to follow the attribution required by the license. Butterick is also one of the lawyers on that case.