Overview
- Elsevier, Cengage, Hachette, Macmillan, McGraw Hill and author Scott Turow filed the class-action Tuesday in Manhattan federal court alleging Meta trained its Llama models on millions of unlicensed books and journal articles taken from pirate sites.
- The complaint says Mark Zuckerberg ended licensing talks in early 2023 and approved using shadow libraries such as LibGen and Sci-Hub to build training datasets.
- The plaintiffs also allege Meta stripped copyright management information, masked its IP address when accessing pirate sites, and kept training sources secret to conceal the copying.
- They seek class certification, damages, court orders to stop the practices, an accounting of training sources, and destruction of allegedly infringing copies, citing examples of Llama reproducing textbook passages and mimicking named authors’ styles.
- Meta says it will fight the case and argues AI training can be fair use, a question now producing mixed rulings in U.S. courts after a 2025 win for Meta in one authors’ case and Anthropic’s $1.5 billion settlement with writers last year.