Anthropic Scores Partial Victory in Copyright Case Over AI Training Data

Athrophica A and Anthropic has won a key legal victory in the Battle of Copyard Rights over how artificial intelligence companies use material protected by copyright to train their models, but the fight is far from the end.

American District Judge William Alsup found that Anthropic is the use of books protected by copyright for training of his AI chatbot Claud qualifies as “honest use” under the US Copyright Law, UA ruling Late Monday.

“Like any reader who strives to be a writer, Anthropic’s lllms trained the works not to run forward and replicate them or replace them – but to turn a hard angle and create something different,” US District Judge William Alsup said in his judgment.

However, the judge also blamed the company Amazon and Google supported for the construction and maintenance of a huge “central library” of the pirate books, calling that part of its operations with a clear violation of copyright.

“No Cutting” from the Copyright Act

The case, brought by authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson last August, The accused anthropically Construction of Claude using millions of pirate books taken from infamous web locations such as the Genesis Library and Pirate Library Mirror.

The lawsuit, seeking compensation and permanent ban, states that the anthrop “built a job with more billion dollars by stealing hundreds of thousands of books protected by copyright,” for the training of Claude, his family AI models.

ALSUP said that AI training can be “extremely transformative”, noting that Claude’s outputs do not reproduce or do not register the works of the author, but create a new text “orthogonal” originals.

Court records reveal that anthropic has taken over at least seven million pirate books, including copies of each author’s works to compile his library.

Internal email revealed that anthropic co-founders sought to avoid the “legal/practice/business syllable” licensing of books, while employees described this goal as the creation of a digital collection of “All Books in the World” that will stay “forever”.

“However, there is no carvery from the AI ​​copyright law,” ALSUP said, noting that the maintenance of permanent libraries is stolen work – even if some are used for training – “destroy the academic publication market” if it is allowed.

The decision of Judge William Alsup is the first essential decision of the US Federal Court to directly analyze and apply the doctrine of honest use, especially to the use of materials protected by copyright for training of generative AI models.

The court distinguished copies used directly for AI training, which was thought to be honest use, and retained pirate copies, which will now be subject to further legal procedures, including potential damage.

Cases of copyright AI

Although several lawsuits have been filed with high-profile cases against opening, meta and other cases are still in the early stages, with suggestions to reject waiting or detection in progress.

Open and Target Both lawsuits are confronted by groups of authors stating that their works protected by copyright have been used without consent to training large language models such as Chatgpt and Llam.

New York Times Sues Openii and Microsoft In 2023, accusing them of using articles about millions of times without permission to develop a tool.

Reddit also recently sued anthropy, stating that the Reddit’s platform scraped Over 100,000 times he trained Clauda, ​​despite claiming he had stopped.