Anthropic Agrees to $1.5 Billion Settlement for Downloading Pirated Books to Train AI

According to legal filing, a case of the use of several million copyrighted books conducted by anthropic authors and publishers has agreed to pay $ 1.5 billion to settle a case Post onlineThe

A federal judge Found in June The use of anthropology million million pirated books was protected under fair use, but it violated copyright laws in a “central library” digital work. The judge ruled that the company’s officials knew that they were downloading pirated work, and a trial was scheduled for December.

This settlement, presented to a federal judge on Friday, will still need final approval but will provide $ 3,000 per book to several thousand authors, New York TimesThe The largest payment of $ 1.5 billion in the history of US copyright law will be paid the largest payment, though the amount paid for the job was often higher. For example, in 2002, a woman from Minnesota paid about $ 9,000 to the downloaded song, a figure was brought down after she ordered her initially payment $ 60,000 per songThe

In a statement to Gizmodo on Friday, the ethnographic ruling from June to the previous judgment that it was involved in fair use by training models with millions of books.

“In June, the District Court issued a breakthrough verdict on AI Development and Copyright Act, identifying the ethnographic view of the AI ​​models as a fair usage,” the Ethnic Deputy Genarian Council said in a statement through email.

“Today’s settlement, if approved, will solve the remaining legacy demands of the plaintiffs. We are committed to developing safe AI systems that people and companies help to increase their power, advance scientific discoveries and solve complex problems, “Sridhar said.

According to legal filing, the anthropologist says that payments will end In all four trangs are bounded in court-appreciated milestones. The first approval of the court settlement of the court will pay the first payment of $ 300 million and $ 300 million within five days of final approval order. Then $ 450 million dollars will be determined with interest within 12 months of the initial order. And finally $ 450 million in the year after that.

Anthropologists, which were recently valued $ 183 billionStill faces the litigation case of companies like RedditWhich has made an agreement In the early 2024 To train Google its AI models on the platform content. And authors still have active cases against other large technology companies like Open, Microsoft and Meta.

The verdict from June has explained that anthropological training of AI models with copyrighted books will be considered as fair use under US copyright law because nobody can read “all modern day classic” and imitate them, which will be protected:

… The creative elements of the work given are not reproduced to the public, not even a writer’s identifiable expressive style … Yes, Claud’s grammar, composition and style outputs that the underlying LLMs eliminate thousands of work. But if someone reads all modern -day classics because of their exceptional expression, they memorize them and then imitate their best writing mix, will it violate the copyright law? Of course not.

The verdict says, “Any reader’s ambitious, ethnographic LLMS has been trained to move forward and replicate them – but to turn a strong angle and create something else,” said the verdict.

Under this legal theory, all the organizations needed to be done to do it legally to training its models. It was a book that is a less than $ 3,000 per book per book. However, as a note as a New York Times, this disposal will not set up any legal instances that can determine the future cases because it is not going to justice.