California Court Grants Final Approval of US$ 1.5 Billion Settlement in Copyright Class Action Against Anthropic

California Court Grants Final Approval of US$ 1.5 Billion Settlement in Copyright Class Action Against Anthropic

The United States District Court for the Northern District of California granted final approval, on July 20, 2026, to the settlement entered into in the class action Bartz et al. v. Anthropic PBC, brought by authors and other copyright holders of books against the artificial intelligence company Anthropic. The settlement provides for the payment of US$1.5 billion to copyright holders of works allegedly obtained by the company through pirated digital libraries. According to the Court’s decision, this is the largest settlement ever reached in a class action involving copyright claims.

The litigation focused on the manner in which Anthropic obtained the books used in the development of its artificial intelligence models. Plaintiffs alleged that the company downloaded digital copies of copyrighted works from pirated digital libraries, such as LibGen and PiLiMi. During the course of the proceedings, the Court found that this allegation represented the legally strongest aspect of the claims.

The action was resolved through a settlement approved by the Court, which entered final judgment in accordance with the terms of the settlement. However, that decision did not result from a definitive adjudication of the copyright infringement allegations asserted by Plaintiffs. Accordingly, the Court did not determine whether Anthropic’s acquisition of the works was unlawful or whether training artificial intelligence models using copyrighted works constitutes fair use, a doctrine under U.S. law that, under certain circumstances, permits the use of copyrighted works without authorization from the copyright holder.

The settlement was entered into in order to avoid the risks associated with continuing the litigation. Had the case proceeded and the jury found that fair use applied or determined that there was insufficient evidence of unlawful copying, Plaintiffs could have concluded the action without receiving any monetary recovery.

By joining the settlement, copyright holders release claims they could have asserted against Anthropic relating to the past acquisition and copying of their works used as inputs for the development of artificial intelligence models, provided that such conduct occurred before August 25, 2025. This release, however, does not encompass other potential violations. Copyright holders remain permitted to judicially challenge any damages arising from responses generated by artificial intelligence models (outputs), as well as conduct undertaken by Anthropic after August 25, 2025.

The settlement provides that the US$1.5 billion fund deposited by Anthropic to resolve the litigation will be distributed exclusively to copyright holders whose works are included in the Works List, the official list of books covered by the settlement. Each eligible work is estimated to receive, on average, approximately US$3,000, an amount equivalent to four times the minimum statutory damages of US$750 provided under U.S. law for ordinary copyright infringements.

To be included in the Works List, a work had to satisfy the eligibility criteria established by the Court, including: (i) having a valid registration with the U.S. Copyright Office; (ii) having a commercial identifier, namely, an ISBN (International Standard Book Number), used by commercial publishers, or an ASIN (Amazon Standard Identification Number), used by Amazon to identify products sold through its platform; and (iii) having been registered with the U.S. Copyright Office within five years after its first publication and before its acquisition by Anthropic or, alternatively, within three months after its first publication.

In addition to monetary compensation, the settlement requires Anthropic to destroy all files obtained through pirated digital libraries, as well as any copies derived from those files. Finally, the Court rejected measures such as the establishment of a mandatory future licensing regime, attribution of the source of works in responses generated by the models, destruction of already-trained models, and a general prohibition on the use of digitized books, finding that such issues exceeded the scope of the class action.

In Brazil, agreements involving the use of copyrighted works in the context of artificial intelligence are also beginning to emerge. In May 2026, Folha de São Paulo and UOL entered into the first agreement in the country with OpenAI to license news content for use in ChatGPT. As a result, the lawsuit filed by Folha de São Paulo in 2025, in which the newspaper alleged the unauthorized use of its content by the platform, was dismissed without a decision on the merits. Under the agreements, OpenAI is authorized to use the news content produced by both media outlets to generate summaries with attribution and links to the original publications, as well as to train its artificial intelligence models. According to the information disclosed by the parties, the license is limited to the editorial content produced by Folha de São Paulo and UOL and does not cover readers’ comments or any other user data. The financial terms of the agreements have not been disclosed.

The decision can be accessed through the link: Order Granting Final Approval of Class Action Settlement — Bartz et al v. Anthropic PBC, Case No. 3:24-cv-05417-AMO (N.D. Cal.)

The article about the agreement between Folha de S.Paulo and OpenAI is available at the following link: Folha de S.Paulo–OpenAI Agreement

Note: For quick release, this English version is provided by automated translation without human review.