The U.S. Federal District Court for the Northern District of California has preliminarily approved a settlement reached by Anthropic, developer of “Claude” over its use of pirate copies of copyright protected books, to train it’s AI models. The settlement was reached in a class action suit filed by authors Andrea Bartz, Charles Graeber and Kirk Wallace, after Anthropic trained its AI models using copyright protected works of authorship by thousands of authors, according to the initial complaint.
The case is an important milestone concerning AI and copyright law, because of a summary judgment issued by the federal district court on June 23. Judge Wiiliam Alsup found that Anthropic’s use of copyright protected works for AI training qualifies as “fair use” under U.S. copyright law, and therefore may not be regarded as infringing use, in and as of itself. The federal court’s reasoning in this regard is may be described to say that an AI training on copyrighted works in order to better generate texts, is no different than a human author reading a book to be inspired by other authors.
This reasoning is consistent with other disputes pending before US federal district courts. In The New York Times v. OpenAI the Times alleges that millions of its articles were copied without permission to train ChatGPT. The federal court for the south district of New York allowed OpenAI defense that the use is transformative and protected by “fair use”. Similarly, in Tremblay v. OpenAI (N.D. Cal. 2024), authors including Sarah Silverman claimed their books were misused to train AI models. The court dismissed some claims, including claims under the DMCA, but allowed others to proceed, suggesting that fair use may apply to AI training, although the scope remains unsettled. These coherent rulings by several federal district courts solidify the approach that AI training use is not copyright infringement, provided copies of such works were lawfully obtained.
Nevertheless, judge Alsup refused to grant Anthropic summary judgment concerning the potential infringement resulting from the AI training use of copies unlawfully obtained by Anthropic. It seems that approximately 465,000 of the copyrighted works used to train the AI model, were downloaded from pirate libraries without payment to copyright holders. Judge Alsup found this may well be copyright infringement, as indicated by the conclusion paragraph of his June 23 ruling:
“This order grants summary judgment for Anthropic that the training use was a fair use. And, it grants that the print-to-digital format change was a fair use for a different reason. But it denies summary judgment for Anthropic that the pirated library copies must be treated as training copies. We will have a trial on the pirated copies used to create Anthropic’s central library and the resulting damages, actual or statutory (including for willfulness). That Anthropic later bought a copy of a book it earlier stole off the internet will not absolve it of liability for the theft but it may affect the extent of statutory damages.”
Anthropic definitely “got the message” and went on to negotiate a settlement. While the settlement amount of $1.5 Billion is enormous, it pales in comparison to the potential liability Anthropic faced at trial, of up to $1 trillion in statutory deranges. It seems that to some extent the monumental scope of the copyright infringement served to decrease the amount of compensation as a result of increasing the complexity of the matter: after previously criticizing the suggested settlement amount as too low, ($3,000 for each of book title used), last Thursday judge Alsup described the deal as “fair”, also considering that administering claims thousands of stakeholders will is a highly complicated legal endeavor. It seems the settlement amount establishes a minimum benchmark for AI copyright disputes.
Together, the abovementioned cases show that U.S. courts are beginning to recognize training AI models on copyrighted works as potentially transformative fair use, but only when the materials are lawfully acquired.
A definitive ruling from Supreme Court or even circuit courts has not yet been issued, leaving the legal framework uncertain and AI companies exposed to ongoing risk. Nevertheless, the district courts’ ruling provide some idea of where copyright law is going where it concerns AI. While the courts recognize the necessity of allowing the use of copyrighted works in order not to stifle the progress of AI technology, there is an effort to set some restraints on the new “wild horse” of the tech world. These signals by the courts may indicate to AI developers that while use of lawfully obtained – paid – copies, is permissible reliance on pirated data may expose companies to substantial liabilities. That understanding could encourage a shift toward large‑scale licensing arrangements and shape the type of licenses required and the costs negotiated.
**The purpose of this circular is to present changes and innovations in the law and it does not constitute legal advice for action in a specific event. We will be happy to assist you with any clarification and/or expansion required. Intellectual Property Department.
