In a first-of-its-kind initiative, Dutch publishers have joined forces with research institute TNO to develop GPT-NL, an ethical AI language model built on legally sourced Dutch-language data. This collaboration aims to set a global benchmark for transparent, copyright-respecting AI training, offering a responsible alternative to large, international AI systems.
The project brings together over 30 national and regional news publishers, including NDP Nieuwsmedia and ANP, who are contributing archived news articles to double the volume of high-quality training data. Contributors are being compensated, ensuring fair use and protecting intellectual property. GPT-NL, trained on 20 billion Dutch-language tokens, is designed to handle real-world applications like summarisation, information extraction, and natural language processing, while adhering to European regulations such as the AI Act.
Unlike traditional AI training practices, GPT-NL incorporates strict safeguards to prevent unauthorized content extraction or reuse when deployed. This not only builds public trust but also helps publishers maintain control over their intellectual assets.
The initiative is a non-profit collaboration between TNO, NFI, and SURF, with the goal of fostering responsible AI innovation that benefits both media stakeholders and the wider economy. By prioritising legal integrity, collaboration, and transparency, GPT-NL could reshape how AI models are developed, setting a new ethical standard for the industry.
This milestone positions the Netherlands as a leader in ethical AI development, offering a blueprint for other nations seeking to balance AI innovation with copyright protection and public trust.
