Authors and publishers whose works were downloaded from unauthorized online libraries may receive about $3,000 per book under what a San Francisco federal court called the largest copyright class-action settlement in history.
By yourNEWS Media Newsroom
A federal judge has granted final approval to a $1.5 billion class-action settlement requiring artificial intelligence company Anthropic to compensate authors and publishers whose copyrighted books were allegedly obtained from unauthorized online libraries.
The San Francisco court described the agreement as the largest copyright class-action settlement in history, according to Mashable.
The lawsuit was brought on behalf of writers who alleged that Anthropic illegally downloaded their books from LibGen and PiLiMi, two online repositories associated with pirated material.
The company then used those downloaded files to build a large internal collection of books, according to the allegations.
The central legal dispute focused on how Anthropic acquired the copyrighted works rather than whether using those books to train artificial intelligence models was lawful.
An earlier ruling repeatedly cited in the final settlement order concluded that training Anthropic’s AI systems on copyrighted material constituted fair use.
That distinction separated the company’s use of the books for model training from its alleged decision to obtain copies through unauthorized sources.
Under the approved settlement, authors and publishers whose pirated works were included in Anthropic’s library may submit claims worth approximately $3,000 for each qualifying book.
That amount is about four times the typical minimum recovery in copyright infringement cases, according to the information provided.
More than 440,000 books have already been claimed.
The final amount paid for each title may depend on the settlement’s claims process and the number of eligible works ultimately approved.
Anthropic must also delete the copyrighted material it downloaded from the unauthorized libraries.
The deletion requirement addresses the continued possession of the allegedly pirated files even though the earlier court ruling treated the use of copyrighted material for AI training as fair use.
The case therefore draws a legal line between acquiring copyrighted works unlawfully and using lawfully possessed material to train an artificial intelligence model.
The authors did not prevail on a claim that Anthropic’s training process itself automatically infringed copyright.
They instead secured the settlement based on allegations that the company built its book collection using files downloaded without permission from LibGen and PiLiMi.
The $1.5 billion agreement resolves the class-action claims without requiring a trial on every disputed issue.
Final judicial approval means the court determined that the settlement was sufficiently fair, reasonable and adequate for the affected class.
The agreement provides a compensation process for authors and publishers while requiring Anthropic to remove the downloaded material at issue.
The ruling could influence future disputes involving artificial intelligence companies that assemble large collections of books, articles, images and other copyrighted material for model development.
AI companies have argued that training systems on copyrighted works can qualify as fair use because models analyze patterns rather than reproduce the original works in their entirety.
Authors, artists and publishers have countered that companies should not be permitted to obtain or use their creations without authorization or compensation.
The Anthropic case did not fully resolve that broader disagreement.
Instead, the earlier ruling supported the legality of AI training under fair-use principles while the settlement imposed substantial consequences for obtaining the underlying works from alleged piracy sources.
That outcome suggests that companies may face different legal risks depending on both the purpose for which copyrighted material is used and the method by which it is acquired.
The record settlement also demonstrates the financial exposure that can arise when hundreds of thousands of individual works are included in a single class action.
With more than 440,000 claimed books and estimated payments of approximately $3,000 per title, the compensation structure accounts for most of the settlement’s total value.
The submitted information did not identify the date by which authors must file claims, when payments will begin or whether Anthropic admitted wrongdoing as part of the agreement.
It also did not specify how the company will demonstrate that the downloaded files have been deleted.
The settlement nevertheless brings one of the most closely watched copyright disputes involving generative artificial intelligence to a conclusion.
Anthropic can continue relying on the earlier ruling that model training constituted fair use, but it must pay authors and publishers for eligible books allegedly obtained through unauthorized libraries and remove those files from its collection.
For copyright holders, the agreement provides an unusually large per-work recovery and establishes a record for class-action copyright settlements.
For the artificial intelligence industry, the case underscores that a court may view model training and the acquisition of training material as two separate legal questions.
Original article
Comments