Artists suing AI companies are beginning to get answers from courts after years of claims that generative AI builders took books, images and music without permission. The cases target companies including Anthropic, Meta, Google, Stability AI, Midjourney, DeviantArt, Runway AI, Suno and Udio, and the early results are mixed enough to annoy both sides.
The fight widened after The Atlantic published a searchable database of works used in AI training. Author Kirk Wallace Johnson found his books, including The Feather Thief and The Fishermen and the Dragon, in the data, according to The Verge. Johnson, who said those books took years to research and write, contacted Susman Godfrey, the firm representing authors in litigation against Anthropic.
Why are artists suing AI companies?
The mechanism is not mystical. AI companies build models by ingesting large datasets so software can identify statistical patterns in text, images or sound, then generate similar outputs. Artists argue that copyrighted work was copied into those datasets without consent, payment or any realistic way to refuse.
Illustrator Sarah Andersen was among the first prominent visual artists to sue. Andersen, Karla Ortiz, Kelly McKernan and other artists filed a class action in January 2023 against Stability AI, Midjourney, DeviantArt and Runway AI. The case followed the release of Stable Diffusion and Midjourney and is still moving through court, according to The Verge.
Musicians are pressing a separate theory against Google. Sam Kogon and other independent musicians accuse Google of using YouTube data and Content ID systems to train its Lyria and ProducerAI music tools in violation of YouTube’s own terms. Google has moved to dismiss the case, arguing in its filing that YouTube’s terms give it broad rights to reproduce, distribute and prepare derivative works. Google spokesperson Jack Malon told The Verge that the company uses YouTube uploads to improve YouTube and Google products, including machine learning and AI applications.
Can AI companies use copyrighted work as fair use?
The answer depends on how the work was obtained and what the court thinks the model builder did with it. In Bartz v. Anthropic, Judge William Alsup found that Anthropic violated copyright law by using pirated ebooks downloaded from the internet to train Claude, but also ruled that scanning legally purchased secondhand books for training under Anthropic’s Project Panama was fair use because it was transformative.
That split ruling produced a concrete artist win. Anthropic agreed to a $1.5 billion settlement, described by The Verge as the largest ever in a copyright case, and agreed to destroy its collection of pirated ebooks. Novelist Andrea Bartz, the lead plaintiff, told The Verge she viewed the settlement as the first major victory for creatives against a large AI company, while disagreeing with the fair-use portion of Alsup’s ruling.
Meta has had some success narrowing the claims against it. Richard Kadrey, Sarah Silverman, Christopher Golden and other writers sued Meta over the alleged use of their books to train Llama. The judge dismissed many claims after finding insufficient evidence of market harm, but narrower copyright claims tied to alleged use of pirated materials remain active, according to the case record cited by The Verge.
The cases have not produced a single rule for AI training. They have produced something more useful: judges are now separating legally acquired material, pirated datasets, platform terms and market harm instead of accepting the industry’s hand-wavy claim that training is automatically fair use. For artists trying to keep a living market for human work, that distinction is the fight.
This story draws on original reporting from The Verge.
