The legal maze of training AI on copyrighted materials

A recent ruling by Judge William Alsup found that Anthropic's use of copyrighted books to train its AI models was lawful, despite ordering the company to pay a $1.5 billion copyright settlement for accessing illegal shadow libraries. Cathy Gellis, an intellectual property attorney, argued that the decision favors AI companies, noting that the fine was trivial compared to Anthropic's projected revenue. Copyright law, last updated in 1976, struggles to address modern AI practices, leading to inconsistent judicial interpretations. Jason Henderson, a senior attorney, highlighted the uncertainty in fair use determinations, emphasizing that courts tend to view AI training as unfair use if it directly competes with the original content. In a similar case, Judge Stephanos Bibas ruled that Ross Intelligence's use of Thomson Reuters' content was not transformative, reinforcing the idea that AI training for competitive purposes is often deemed unlawful. Gellis also pointed out the distinction between AI training and AI-generated content, citing the Thaler v. Perlmutter case, where fully AI-generated works are not copyrightable. As AI companies face ongoing litigation, the legal landscape remains unclear, with early rulings shaping future outcomes but no definitive resolution in sight.