Six prominent authors including Pulitzer Prize winner John Carreyrou filed individual lawsuits on December 23 against OpenAI, Google, Meta, Anthropic, xAI, and Perplexity AI, seeking $150,000 in statutory damages per work from each defendant—potentially totaling $900,000 per book. The plaintiffs opted out of a $1.5 billion Anthropic class action settlement they characterize as serving AI companies rather than creators, escalating copyright battles over unauthorized use of pirated books for model training.
The lawsuits filed in the Northern District of California allege AI companies systematically downloaded copyrighted books from pirate libraries including LibGen, Z-Library, and OceanofPDF to train large language models without permission, licensing, or compensation. Carreyrou, known for exposing the Theranos fraud scandal in his book "Bad Blood," leads a group of writers arguing that earlier settlements fail to address the fundamental issue of training on stolen intellectual property.
Settlement Backlash Drives Independent Actions
The new filings represent direct rejection of the proposed Anthropic settlement offering approximately $3,000 per eligible work to class members. While a federal judge ruled in June that AI training on copyrighted material constitutes fair use if books were acquired legally, the same ruling found that pirating books was not fair use—a distinction the settlement addresses through monetary compensation without prohibiting future training on legitimately obtained works.
The plaintiffs argue this resolution structure benefits AI companies disproportionately by eliminating thousands of high-value claims at minimal cost. Their complaint states the settlement "seems to serve [the AI companies], not creators," allowing defendants to resolve massive willful infringement allegations without meaningfully constraining their business practices or compensating rightsholders for the commercial value their works enabled.
By filing individually rather than joining class actions, the authors preserve claims for maximum statutory damages under the Copyright Act. The $3,000 settlement payment represents just 2 percent of the $150,000 statutory ceiling available for willful infringement, which plaintiffs argue more accurately reflects the scale of unauthorized use and commercial value generated through training on their works.
First Lawsuits Against xAI and Perplexity for Training
The filings mark the first copyright lawsuits against Elon Musk's xAI over training processes and the first suit by authors against Perplexity AI. This expansion beyond OpenAI, Google, Meta, and Anthropic signals authors' intention to hold the broader industry accountable for training data practices.
Legal documents describe a "clear cycle of infringement: Pirated Acquisition, Model Training, and Commercial Monetization" where literary works "anchor multibillion-dollar product ecosystems without compensation." More than 50 copyright cases between intellectual property holders and AI developers now proceed through federal courts, with OpenAI facing 14 total copyright cases—the most legal pressure in the industry.
ClaimsHero Controversy Complicates Settlement Process
The individual lawsuits follow controversy around ClaimsHero, a website encouraging authors to opt out of the Anthropic settlement to pursue separate litigation. Judge William Alsup, who presided over the Anthropic case, called ClaimsHero's communications "a fraud of immense proportions" and ordered the organization to alter "misleading" claims after class action lawyers sought to stop opt-out solicitation.
Authors have until January 16 to decide whether accepting the Anthropic settlement or pursuing independent claims. The deadline creates pressure as plaintiffs weigh guaranteed modest payments against uncertain prospects of substantially larger awards through individual litigation potentially lasting years with no compensation guarantee.
Industry Implications Extend Beyond Damages
Beyond immediate financial stakes, the lawsuits challenge whether AI companies can continue training on vast unlicensed datasets or must negotiate licensing agreements with rightsholders. Successful plaintiff outcomes could force expensive content licensing frameworks fundamentally altering AI economics through intellectual property compliance requirements.
If courts impose substantial damages for piracy-based training, AI companies face pressure to verify content provenance and potentially limit datasets to licensed or public domain materials—constraints that could affect model capabilities and competitive positioning across the industry.
