French organisations representing publishers and authors said Wednesday they were launching legal action against Facebook owner Meta, after their books were used to train generative AI applications.
The three groups – publishers’ outfit SNE and authors’ and composers’ groups SGDL and SNAC – complained in a statement of “massive use of copyrighted works without authorisation from their authors and publishers” by the American company.
“We have established the presence of many works published by SNE members in the body of data used by Meta,” SNE chief Vincent Montagne said in the statement.
Meta has acknowledged using a database, Books3, containing the full texts of around 200,000 books including some in French to train its Llama large language model.
In a separate US court case launched by authors, the company admitted last year to using the database until 2023, claiming that the AI training constituted “fair use” of the copyright-protected books.
French publishers and authors have not publicly communicated an estimated value of the harm to them by Meta.
Their case at the Paris judicial court “should lead to a serious desire emerging on the part of AIs to take the creative industries into account”, SGDL head Christophe Hardy said.
He called on AI developers to “respect the legal framework and, where relevant, find compensation for the use of works that feed into” the technology.