We have been contacted by a number of our authors following a recent piece in The Atlantic which references Meta’s admission that the pirate network Library Genesis or ‘LibGen’ has been used as training data for Meta’s large language models (LLMs).
Training data fuels how LLMs learn to mimic language, style, and ideas. When copyrighted materials are used unlawfully, it raises serious legal and ethical questions about authorship, compensation and creative control.
Our position
We share the concerns and frustrations expressed by our authors and are fully committed to protecting copyright and other intellectual property rights. We do not permit works we have published to be used for the training of AI models and regard any such unauthorised use as a breach of copyright and/or contract.
As members of the Publishers Association, we fully support their ongoing work to try and address this vital issue. Working closely with their counterparts and author groups around the world, they are urgently trying to address the unauthorised, large-scale use of copyright-protected works in the training of generative AI models.
The Publishers Association have written to Meta, and around 50 other AI developers, to put them on notice that publishers do not authorise or otherwise grant permission for the use of copyright-protected works in this way.
Bonnier Books UK responded directly to the government’s recent consultation on copyright law reform and AI, to outline our concerns. We also supported the Publishers Association’s submission.