Anthropic has sidestepped the biggest potential copyright headache of many it faces from content creators.
In a court filing in California federal court Tuesday, the company revealed it settled a class-action lawsuit with a trio of authors who claimed the artificial intelligence (AI) startup illegally downloaded and copied their books. Under terms of the settlement, announced on Sept. 5, Anthropic agreed to pay the authors $1.5 billion, or about $3,000 for each of an estimated 500,000 books covered by the settlement.
“As best as we can tell, it’s the largest copyright recovery ever,” said Justin Nelson, a lawyer for the authors who last year sued Anthropic for training its AI models on pirated versions of their works. “It is the first of its kind in the AI era.”
The dispute, like many others in court, centered on whether AI was covered by the fair use doctrine, intellectual property law that lets creators build upon copyrighted works without a license. Federal judge William Alsup in June agreed with Anthropic that its process of training models on copyrighted works was “exceedingly transformative,” but he nonetheless ordered a December trial to let the authors — Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson — press the company on how it obtained their books.
Anthropic said it legally bought paperback versions of the authors’ books, scanned and digitized their material, and then destroyed the originals. Yet it also illegally downloaded seven million books to accelerate training of Claude and kept those copies. Though it consequently purchased the mountain of books, it failed to shed its liability in doing so belatedly, Alsup ruled.
Had it gone to court and lost, Anthropic could have been on the hook for up to $1 trillion in damages by some estimates.
The settlement of its most high-profile copyright dispute hardly means that Anthropic is out of the legal woods. It and other AI companies face a torrent of legal disputes over the use of copyrighted material.
OpenAI is being sued by publisher Ziff Davis after it discovered that ChatGPT had “relentless reproduced exact copies and created derivatives” of articles in PCMag. Perplexity AI is being sued by media conglomerate News Corp. for repackaging its content. And the New York Times has taken legal action against OpenAI and Microsoft Corp. for the use of Times articles to train GPT large language models without its permission.
