Court Draws the Line: AI Training Is Fair Use, But Pirated Libraries Cost $1.5 Billion

A U.S. judge approved a record $1.5 billion settlement over Anthropic's pirated book library while affirming that training AI on published works is fair use.

A U.S. federal judge has given final approval to a $1.5 billion settlement between Anthropic and a class of authors and publishers, reported as the largest payout in a copyright case. It is also the first settlement to close among the many copyright suits now being litigated against AI companies. The details are attributed to Reuters reporting, dated 21 July 2026.

The ruling splits cleanly. The court found that training AI on books qualifies as fair use under copyright law. What it did not allow was piracy: Anthropic kept 7 million pirated books in a central library, and that copying violated the rights of the authors and publishers involved. The settlement works out to a little over $200 per title. According to Anthropic deputy general counsel Aparna Sridhar, the deal was reached in 2025 after the fair-use ruling, and more than 91% of covered rightsholders claimed their share.

For open knowledge, the distinction matters. Affirming training as fair use keeps the legal path open for building models on published material, which lowers the cost of entry for research and open development. Penalizing the pirated library says the path runs through legitimate acquisition, not torrented archives. Both halves push toward a commons where models can be built without wholesale theft.

Watch the cases that did not settle. Some groups opted out and filed separately, and separate allegations name Nvidia's NeMo download scripts and Meta's reported torrenting of 82TB of books. Those allegations remain unproven, and other courts may draw the fair-use line differently.

Source: Tom's Hardware