Sarah Silverman and Authors Sue Meta and OpenAI for Copyright Infringement Over Book Training Data

In July 2023, comedian and author Sarah Silverman, along with writers Christopher Golden and Richard Kadrey, filed separate but related class-action lawsuits against Meta Platforms and OpenAI in the U.S. District Court for the Northern District of California. The lawsuits alleged that both companies' AI language models - ChatGPT/GPT-3.5/GPT-4 from OpenAI and LLaMA from Meta - were trained on copyrighted books without permission from the authors or publishers. The plaintiffs claimed that their works were obtained from Library Genesis (LibGen), a notorious "shadow library" website that hosts millions of pirated books and academic papers. The lawsuits alleged that the defendants used these unauthorized copies as training data for their large language models, enabling the AI systems to generate summaries and analyses of the copyrighted works. As evidence, the plaintiffs demonstrated that ChatGPT could produce accurate summaries of their books when prompted, suggesting the models had been trained on the full text of their works. The legal filings argued that this constituted direct copyright infringement, vicarious copyright infringement, and violations of the Digital Millennium Copyright Act (DMCA). The authors sought monetary damages, including profits derived from the alleged infringement, as well as injunctive relief to prevent further unauthorized use of their works. The cases were structured as class actions, potentially representing thousands of authors whose works may have been similarly used without permission. Both Meta and OpenAI filed motions to dismiss the lawsuits, arguing that their use of copyrighted material constituted fair use under copyright law. The companies contended that training AI models on copyrighted texts for the purpose of learning patterns and generating new content fell within the bounds of transformative fair use. They also argued that the plaintiffs failed to demonstrate concrete harm or that the AI outputs constituted infringing derivative works. The litigation represents a landmark test case for how copyright law applies to AI training data, with implications for the entire generative AI industry. The outcomes could establish important precedents regarding whether using copyrighted works to train AI models requires explicit permission from rights holders, or whether such use qualifies as fair use. As of late 2023, the cases remained in the discovery phase, with both sides gathering evidence about training data sources and methodologies.