Judge Rules AI Companies Can Use Some Copyrighted Works to Fuel Their Sludge

This week, a federal judge gave AI companies a major victory, which potentially established a legal precedent for the industry to fold the documents protected by copyright to train its major language models.

Anthropic, the great IA company supported by Amazon, participated in a legal battle with a group of writers and journalists who continued the company last summer And accused her of illegally used his work to form the company’s flagship chatbot, Claude. The legality of the entire business model of the AI ​​industry long depends on the question To know if it is a kosher to increase large amounts of data protected by copyright from all over the web, then transmit it into an algorithm to produce “original” text. Anthropic argued that his use of the work of writers falls under equitable use And is therefore legal. This week, the federal judge presiding over the case William Alsup, partially accepted.

In His decisionAlsup said that by forming his LLM without the authorization of the authors, Anthropic did not reach the material protected by copyright because the work he produced was, in his eyes, original. He said that business algorithms had …

“… no reproduced to the public creative elements of a given work, or even an identifiable expressive style of an author … Yes, Claude has released grammar, composition and style that the underlying LLM distilled from thousands of works. But if someone should read all the modern classics because of their exceptional expression, memorize them, then imitate an extension of their best writing, which would violate copylight?

Alsup’s decision starts a little from the dispute of writers, which has accused anthropogenic of human expression “extraction” and ingenuity for business benefits. This decision is only the opinion of a judge, but criticism fear that he can easily establish a precedent for other legal decisions across the country. AI companies have been pursued dozens of times by creatives on similar reasons.

Although Alsup’s decision can report wider victories for the AI ​​industry, this is not exactly what you would call a victory for Anthropic. Indeed, Alsup also judged that the specific way in which Anthropic has drawn some of the documents protected by copyright for its LLM – by downloading more than 7 million pirated pounds – could be illegal and would require a separate test. “We will have a test on the pirated copies used to create the central library of Anthropic and the resulting damage,” wrote Alsup. “This anthropic later bought a copy of a book [that] He stole the Internet earlier will not get him responsibility for theft, but this can affect the extent of statutory damages. »»

When he was contacted to comment by Gizmodo, Anthropic provided the following declaration: “We are happy that the Court recognized that the use of” work to form the LLM was transformer – spectacularly thus “. In accordance with the purpose of copyright to allow creativity and promote scientific progress, “ LLMS of Anthropic has trained on the works not to run in advance and reproduce them – but to turn a hard corner and create something different. ”

Alsup has chaired several leading cases involving large technological companies, including Uber, Doordash and Waymo. More recently, Alsup has ordered the Trump administration To reintegrate thousands of probation workers drawn which were expelled by the DOGE initiative of Elon Musk.