Mark Zuckerberg, CEO of Meta, allegedly allowed the team behind the Llama AI models to use a dataset of illegally obtained e-books and articles, known as LibGen, for their training courses. This is according to documents recently filed in the lawsuit Kadrey v. Meta .
Copyright violations
LibGen is known as a platform that provides access to thousands of copyrighted works. It has been sued and convicted multiple times for copyright violations. Despite warnings from Meta’s own AI experts about the legal risks, according to the indictment, Zuckerberg personally consented to the use of this dataset.
In addition, Meta is accused of manipulating the data. The team allegedly removed copyright notices and metadata to prevent users from noticing that the models were trained on protected works. Meta also allegedly downloaded the data via torrenting – a practice that is not only illegal, but also poses risks to the distribution of the content.
The lawsuit has not yet been decided and for now is limited to older Llama models, but reputational damage has already been done to Meta. One judge called Meta’s attempts to hide details “aimed at avoiding bad publicity” rather than protecting company information.