Dutch Authors Sue Meta Over AI Training Data

Tech Giant Accused of Widespread Copyright Infringement

A growing number of Dutch authors are voicing outrage against Meta, the parent company of Facebook, Instagram, and WhatsApp, alleging unauthorized use of their works to develop artificial intelligence chatbots. The authors claim their literary creations were used without permission to train Meta’s sophisticated language models.

Allegations of “Mafia Practices”

Court documents originating from the United States reveal that Meta allegedly accessed dozens of terabytes of text data from an illicit online repository known as Libgen (Library Genesis). This vast collection, reportedly containing approximately 7.5 million books and 81 million research articles, served as the foundation for Meta’s AI chatbot development. Crucially, court filings indicate that texts from Dutch authors were also included in this database.

Bestselling author and journalist Sacha Bronwasser expressed her dismay, stating, “I feel robbed.” Fellow writer and director Philip Huff echoed this sentiment, describing Meta’s actions as “absurd.” He added, “But because they are so big, they get away with it. It is just classic colonialism: a very large entity takes what it wants.”

An analysis by *Nieuwsuur* found thousands of books by Dutch authors within the Libgen database, which has also reportedly been utilized by other major tech firms like Anthropic and OpenAI. For instance, more than half of the 49 best-selling books by Dutch authors in 2023 were allegedly present in the database when Meta accessed it. Some authors, such as novelist Arthur Japin, had as many as twenty of their titles listed.

Authors’ Association Calls for Action

The Dutch Authors’ Association, a professional organization, has urged authors who feel wronged to come forward. The association has already received over 600 responses, according to Anja Sicking, chairperson of its Literary Authors section. She condemned the tech companies’ conduct as “Mafia practices.”

“You may not just use copyright work of others if you do not have permission for that.”

—Charlotte Meindersma, AI-Jurist

Legal Battles and Fair Use Doctrine

Popular chatbots like ChatGPT are trained on extensive text datasets, enabling them to recognize patterns and generate human-like text through artificial intelligence. Meta’s own language model, now accessible to Dutch users on Facebook and WhatsApp, operates on a similar principle. However, companies typically do not seek explicit permission from authors for this training data, leading to a surge of lawsuits.

Recently, American judges dismissed lawsuits filed by authors against Meta and Anthropic, citing the “fair use” doctrine. The judges reasoned that the companies’ language models create new works, and therefore, the use of existing texts constituted fair use. This legal precedent is crucial in the ongoing debate over AI and copyright.

However, AI and marketing lawyer Charlotte Meindersma pointed out that the “fair use” principle is not recognized in the Netherlands or Europe. “And so what Meta does is not allowed here,” she stated. “You may not just use copyright work of others if you do not have permission for that.” Meindersma clarified that while an exception exists for using large text volumes for scientific research, it does not apply here, as Meta’s purpose was commercial.

Meta, in response to *Nieuwsuur*, maintained that its actions are permissible and that its language models foster “incredible innovation, productivity and creativity.” The company declined to answer specific questions regarding its data usage.

EU Regulations and Future Steps

Writer Philip Huff challenged Meta’s claims, arguing that “Innovation is not an excuse for immoral activities.” He further questioned the company’s purported innovation, suggesting that the reliance on millions of books indicates a potential lack of original development.

The Authors’ Association is currently exploring legal recourse. “But that is complicated, because they are very big American parties that you have to fight against,” said Sjors Hofstede.

The European Union is also taking steps to regulate tech companies’ data practices with new legislation. A code of conduct, set to take effect shortly, mandates that companies obtain permission and disclose the data used for training their AI models. Meta has indicated its unwillingness to sign this code. The European Commission emphasized that all companies must comply with the AI Act starting August 2, and non-compliance could result in increased scrutiny from Brussels.