A recent court decision has shed light on how “fair use” applies to generative AI systems – a significant ruling that is likely just the opening salvo of a long legal battle.
Fair use generally permits third-party use of copyrighted material without permission in specific situations, such as quoting a song lyric in a news article. AI companies often cite fair use when defending their use of copyrighted works to train their generative AI models. However, authors and creative industries are challenging these claims.
In a 2024 class-action lawsuit, authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson alleged that Anthropic AI’s chatbot, Claude, was trained on millions of digitized copyrighted books, including their own. Anthropic reportedly purchased and scanned some physical books to feed its model without obtaining permission or paying a fair price.
In June, Senior U.S. District Judge William Alsup in the Northern District of California ruled that the company’s use of the plaintiffs’ books for training constituted fair use. A particularly interesting aspect of Judge Alsup’s ruling was his comparison of AI training to how humans learn. He wrote that everyone reads and then writes new texts, and while payment for initial access to a text may be necessary, requiring payment for every reading or recall would be “unthinkable.” The court also observed that Anthropic’s large language models, “like any reader aspiring to be a writer,” trained on works “not to race ahead and replicate or supplant them – but to turn a hard corner and create something different.”
Judge Alsup noted, however, that Anthropic also “downloaded for free millions of copyrighted books in digital form from pirate sites on the internet” to build its universal library. On this basis he allowed the authors’ piracy complaint to proceed to trial.
Similar lawsuits against AI companies have been filed by prominent authors including Ta-Nehisi Coates, Michael Chabon, and Junot Díaz. On the heels of the Anthropic decision, another judge in the Northern District of California ruled in favor of Meta in a lawsuit brought by 13 authors, including Richard Kadrey and Sarah Silverman, who claimed Meta used pirated copies of their novels to train LLaMA. Meta’s fair use defense held up because the authors could not prove that Meta’s use harmed the market for their original works. However, the court suggested that future cases with stronger arguments could lead to different outcomes.
These rulings may seem like wins for AI companies, but appeals are highly likely, potentially reaching the Supreme Court. While the issue of fair use continues to play out in court, both AI companies and content creators are digging in for the long haul.