Artists file more lawsuits over AI training data, and some win in court

In short: More artists are suing AI companies over how their work was used to train AI systems, and a few cases are starting to produce real legal consequences.

What's going on

More writers, illustrators, and musicians are taking AI companies to court, saying their work was copied and used without permission to train AI. Training data is the material an AI studies to learn patterns, like a student reading a huge stack of books and listening to lots of songs.

Some artists say they found their work listed in public datasets that track what was used to train AI. Writer Kirk Wallace Johnson said his books appeared after The Atlantic published a searchable dataset, and he joined ongoing legal efforts. Visual artist Sarah Andersen and others filed a class action lawsuit in 2023 against companies including Stability AI and Midjourney, and that case is still moving through court.

The lawsuits often focus on copyright, which is the legal right to control copying of your work. Others target contract rules, like platform terms. Musician Sam Kogon and other independent artists are suing Google, arguing YouTube data and systems like Content ID were used to train Google’s Lyria, and that this broke YouTube’s terms. Google has asked the court to dismiss the case and says its terms give broad rights, and a Google spokesperson said YouTube content is used to improve products, including through AI.

What to watch

One major sign of momentum came in Bartz v. Anthropic. The court found Anthropic violated copyright law by using pirated ebooks to train Claude, leading to a $1.5 billion settlement and an agreement to destroy the pirated ebook collection. At the same time, the judge said training on legally bought and scanned books could count as “fair use” in some situations, so the line is still being debated in court.

Source: The Verge AI