Anthropic will pay at least $1.5 billion for its previous use of pirated books, according to the headlines in the settlement that was approved on September 25, 2025, although I have already written about this in earlier articles. This amount represents $3,000 for each of the approximately 500,000 copyrighted works that were alleged to have been downloaded from LibGen or PiLiMi and infringed upon by Anthropic. The parties are currently in the process of compiling the complete list of copyrighted works. If the list surpasses 500,000, Anthropic will provide an additional $3,000 for each additional work.
gov.uscourts.cand.434709.375.0.pdf
gov.uscourts.cand.434709.383.0_3.pdf
The above two links contain the questions posed by the presiding Judge William Alsup, who grants preliminary approval of the $1.5 billion settlement—the largest in copyright history— in the class action filed by Bartz v. Anthropic. – The article “Chat GPT Is Eating the World” emphasises the significance of the settlement and its implications for published authors.
Only past releases of liability are available: It is important to note that the settlement does not protect Anthropic from potential future claims. The plaintiffs have only disclosed claims regarding Anthropic’s prior acquisition, retention, and use of the identified works for internal research and development activities related to AI training and development that took place prior to August 25, 2025. The settlement only covers works identified in the final inventory and does not apply to future conduct or other works owned by class members.
Although the claims at issue in this case were exclusively related to the materials used to train Claude (i.e., the “input” to the LLM), a common issue in many AI-related copyright actions is whether the LLM produces output that infringes copyrighted works. The settlement between the parties does not address any allegations of infringement of the outputs of Anthropic’s AI models. This implies that the claims are still actionable if Anthropic’s models produce infringing outputs, even if the outputs infringe on the same identified works.
In addition to the monetary compensation, Anthropic has consented to the destruction of the two libraries that are purportedly home to the pirated works, as well as any derivative copies that originate from those sources, within 30 days of the final judgement or the expiration of any litigation preservation obligations. Anthropics is required to provide written confirmation to class counsel that the destruction has been finalised and that the allegedly infringing materials have been irrevocably removed from its systems.
What does this agreement mean?
This landmark settlement underscores the significance of comprehensive compliance strategies for AI developers and users. The per-work settlement amount of $3,000 is four times the $750 floor for statutory damages and 15 times the amount that the plaintiffs would have received had Anthropic prevailed at trial on its defence of innocent infringement and convinced a court to exercise its discretion to revise the damages to $200 per work.
Consequences for Other Pending Cases Against AI Developers
Other pending copyright actions against AI companies are likely to use the Anthropic settlement as a benchmark. In comparable cases, plaintiffs may use the $3,000 per-work figure as a starting point for damages negotiations, and courts may consider the structure of this settlement when determining remedies. Even though the settlement isn’t a rule that must be followed, AI companies that could face big legal issues and similar claims should expect that plaintiffs will follow the example set by this settlement and push for the identification, separation, and removal of any data sources that violate copyright during negotiations. This requirement may also apply to derivative datasets and any internal versions, thereby increasing the importance of data governance and compliance.
We anticipate exciting times ahead!
Michael.Coyle@lawdit.co.uk
