Anthropic Must Face Pirated Books Trial Despite AI Ruling

    0
    0

    In a significant ruling impacting the artificial intelligence industry, a federal judge determined that Anthropic did not violate copyright law when training its chatbot, Claude, using millions of copyrighted books. However, the company must still undergo a trial concerning the acquisition of these books from illicit online “shadow libraries.”

    U.S. District Judge William Alsup in San Francisco declared that the AI’s extraction of texts from numerous works constituted “fair use” under U.S. copyright laws because it was “quintessentially transformative.” He elaborated that Anthropic’s AI models, like aspiring writers, do not aim to duplicate or replace the sources but strive to create uniquely new material.

    While dismissing an essential claim posed by authors accusing the company of copyright infringement last year, Judge Alsup emphasized that Anthropic’s trial is set for December to address allegations of unauthorized book usage. “Anthropic had no right to engage with pirated copies as its primary resource,” he wrote.

    Authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson filed a lawsuit last summer accusing Anthropic of orchestrating “large-scale theft” via exploitation of the creative expressions and efforts embedded in these works. Books serve as critical data reservoirs—essentially aggregations of billions of meticulously arranged words—needed for crafting advanced AI language models. In the competitive rush to excel in AI chatbot development, numerous tech firms, including Anthropic, have turned to online platforms of illegally obtained books for free access.

    Court documents in San Francisco revealed internal debates among Anthropic employees about the legality of using pirated materials. Anthropic later adjusted its approach by hiring Tom Turvey, a former Google executive involved with Google Books, a digitized library that weathered numerous copyright legal challenges. With guidance from Turvey, Anthropic began procuring books en masse, dismantling bindings and scanning pages before feeding them into AI systems. However, Judge Alsup remarked that this practice did not rectify previous piracy issues.

    “The later purchase of copies does not negate earlier liabilities for theft,” stated Alsup, adding that it might influence the assessment of statutory damages. This ruling might influence related lawsuits against companies like OpenAI, creators of ChatGPT, and Meta Platforms, Facebook and Instagram’s parent company.

    Founded by ex-OpenAI leaders in 2021, Anthropic branded itself as a more responsible, safety-centric developer of AI models designed for composing emails, summarizing documents, and engaging naturally with users. However, last year’s lawsuit criticized Anthropic for undermining its proclaimed goals by using pirated content for AI development.

    Following the court decision, Anthropic expressed satisfaction that the judge acknowledged AI training as transformative and aligned with fostering creativity and scientific advancement, as per copyright’s purpose. The company did not address piracy accusations in its statement, while the authors’ legal representatives declined to comment.