In a groundbreaking decision that is already sending shockwaves through the tech and creative industries, a federal judge has ruled that the act of using copyrighted books to train artificial intelligence is a "quintessentially transformative" fair use.
The ruling came in a high-stakes lawsuit filed by a group of authors against the prominent AI firm Anthropic, creator of the AI assistant Claude. The decision, however, is a split one with a critical distinction: while the training itself is protected, the court found that Anthropic may still be liable for building its digital library with books downloaded from pirate sites.
This landmark order carves out a new legal landscape for AI, suggesting a path forward for innovation while putting the entire industry on notice about where it gets its data. The decision is almost certain to be appealed and will likely set the stage for years of legal battles.
What the Judge Ruled
The case hinged on how Anthropic built its massive internal library and what it did with it. The company amassed millions of digital books in two ways: by purchasing and "destructively scanning" physical copies, and by downloading vast collections from known pirate sites. The court carefully separated these actions.
1. Training AI Is Fair Use This is the bombshell. Judge William Alsup ruled that the process of using copyrighted material to teach a large language model is a transformative act protected under fair use. He argued that the AI, like a human student, reads the material to learn concepts, patterns, and styles, not to regurgitate the original work.
The resulting AI model is not a copy or a substitute for the books themselves. This is a monumental win for AI companies like Anthropic, Google, and OpenAI.
2. Digitizing Owned Books Is Also Fair Use The court also sided with Anthropic on its practice of buying physical books and scanning them into a digital library. The judge reasoned that this was simply a "format change" for legitimately owned copies, allowing for more efficient storage and searchability, and did not constitute infringement.
3. But Piracy Is Still Piracy Here is where the court drew a bright red line. Judge Alsup denied Anthropic's request to have the use of pirated books excused. He ruled that an eventual fair use (training) does not absolve the initial act of theft. Creating a permanent library from stolen works that could have been purchased harms the authors' market.
This part of the case will now proceed to determine damages, a significant victory for the authors.
Specifically, the ruling said:
"This order grants summary judgment for Anthropic that the training use was a fair use. And, it grants that the print-to-digital format change was a fair use for a different reason. But it denies summary judgment for Anthropic that the pirated library copies must be treated as training copies."
Why This Ruling Changes Everything
This decision is the first of its kind to so clearly separate the act of training an AI from the act of acquiring the training data.
For AI companies and those who want to learn AI, it provides a powerful legal shield for the core process of training their models, so long as the final product is transformative and not a market substitute for the original works.
For authors, artists, and creators, it affirms that their copyrights cannot be ignored. AI firms cannot simply scrape the internet's dark corners for free material without consequence. The ruling empowers creators to fight back against the mass acquisition of their work from illegal sources.
What's Next: An Inevitable Appeal
This is just the beginning. While Anthropic won on the central question of training, the financial penalty for using pirated books could still be substantial. The ruling will undoubtedly be appealed by the losing side on each issue, likely sending the case to a higher court and eventually, perhaps, the Supreme Court.
The messy, brilliant, and now legally complex era of generative AI just got its first major landmark. The lines have been drawn, and the fight over the future of creativity and technology is far from over.
Note that the court document from June 23rd, appears online here.
