The UK’s Data (Use and Access) Bill was passed without crucial amendments requiring AI tools to disclose their usage of copyrighted material or allowing copyright holders to opt-out. This raises concerns about whether AI can continuously exploit copyrighted works.

A recent ruling by a federal judge found that Anthropic, the AI company known for its model Claude, did not violate copyright law by utilizing copyrighted content for AI training, as it constitutes ‘fair use.’ However, the issue arises regarding how Anthropic sourced some of that content—specifically, thousands of books available online without proper purchase. Essentially, while AI can use copyrighted material, it cannot engage in piracy.

This legal journey began when authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson filed a lawsuit last summer against Anthropic. They alleged that the company acquired and used pirated versions of their works for training its AI model. There have been internal concerns voiced by Anthropic employees about the utilization of these pirated texts. Although Anthropic later shifted to legally acquiring and digitizing physical books for AI training, the judge’s ruling emphasizes the need to address how the company initially obtained copyrighted works.

Judge Alsup stated, “Anthropic had no entitlement to use pirated copies for its central library.” He described the output of AI models trained on copyrighted material as “quintessentially transformative,” thus not violating fair use laws. Alsup also suggested that the purpose of training AI is not to replicate but to create original content inspired by existing works.