AI Company Anthropic’s Fair Use Ruling and Copyright Violation

A federal court in San Francisco has determined that AI enterprise Anthropic’s utilization of copyrighted literature for training its large language model, Claude, is permissible under U.S. fair use statutes. However, the decision marked a significant issue concerning the company’s storage of millions of pirated books, which has been classified as a copyright infringement. This development sets the stage for a forthcoming trial in December to ascertain potential damages.

Judge’s Insights on Transformative Use

U.S. District Judge William Alsup asserted that Anthropic’s incorporation of texts by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson for AI training was “exceedingly transformative” and aligns with the intent behind copyright laws. The ruling articulated that Anthropic’s large language models (LLMs) are akin to any reader-turned-writer, aiming not merely to reproduce or substitute the original works, but to offer a fresh perspective and innovative outputs.

Criticism of Pirated Book Storage

Despite acknowledging the transformative use, Alsup critiqued Anthropic for maintaining over 7 million pirated books in what he referred to as a “central library.” He concluded that this accumulation is not afforded protection under fair use and infringes on the authors’ rights, even if those specific works were not actively employed in the training process.

Authors’ Claims Against Anthropic

The authors involved, who initiated a proposed class action lawsuit last year, contended that Anthropic had exploited unauthorised versions of their works without permission or remuneration. Anthropic enjoys backing from major technology corporations, including Amazon and Alphabet.

Anthropic’s Reaction

An Anthropic representative expressed approval regarding the court’s partial approval of their case: “We are pleased that the court recognised our AI training as transformative and consistent with copyright’s purpose in enabling creativity and fostering scientific progress.”

Implications for AI Companies

Judge Alsup’s ruling is recognised as a pioneering U.S. opinion that directly examines the concept of fair use in the realm of generative AI, a key defense mechanism employed by firms such as OpenAI, Meta, and Microsoft in ongoing copyright disputes.

While AI companies maintain that training on copyrighted materials facilitates the creation of innovative works and spurs progress, copyright holders argue that such practices erode their potential earnings from original creations. The ruling balances these viewpoints, legitimising transformative AI training whilst condemning the illegal custody of pirated assets.

Potential Damages in Upcoming Trial

The anticipated trial may result in Anthropic facing statutory damages reaching up to $150,000 per work for deliberate copyright infringement. This case is poised to sway upcoming legal confrontations regarding how AI enterprises access and utilise copyrighted materials in developing generative models.