TLDR: Dutch publishers, through the anti-piracy agency BREIN, have successfully compelled Common Crawl, a non-profit web archiving organization, to delete over two million news articles from its database. These articles, sourced from prominent Dutch news websites and digital newspapers, were used without authorization to train various generative AI models. The move highlights growing concerns over copyright infringement and the need for legal licensing in the AI training landscape, with calls for collective compensation for journalists.
In a significant development for copyright in the age of artificial intelligence, Dutch publishers have successfully forced Common Crawl to remove more than two million news articles from its extensive database. The action was spearheaded by BREIN, the Dutch anti-piracy agency, acting on behalf of numerous Dutch news organizations. The removed content originated from well-known Dutch news websites and digital newspapers, which had been scraped and included in Common Crawl’s dataset without explicit permission.
Common Crawl, an American non-profit organization, is renowned for its vast web archives, which it makes freely available to both consumers and businesses. Crucially, these datasets have been a foundational resource for training many generative AI services and large language models (LLMs). Research indicates that a substantial number of major AI models, including Apple’s openELM, Microsoft’s Phi, OpenAI’s ChatGPT, NVIDIA’s Nemo Megatron, Deepseek’s DeepseekV3, and Anthropic’s Claude, have been at least partially trained using Common Crawl data.
The primary objective behind BREIN’s request was to prevent AI services from continuing to train their models on this unauthorized content. Common Crawl has reportedly complied with the demand, initiating the removal of the specified articles. This incident underscores the escalating tensions between content creators and AI developers regarding the use of copyrighted material for AI training.
Publishers are increasingly seeking legally licensed alternatives for AI companies to access their content. Examples include The Guardian’s arrangement with OpenAI, established in February 2025, and The New York Times’ reported deal worth at least $20 million per year, announced in May. These agreements signify a shift towards compensated use of journalistic content for AI development.
In the Netherlands, an initiative called GPT-NL, a large-scale Dutch AI language model, began training in June 2025. This model is notable for being entirely trained on legally obtained data, a collaborative effort by Dutch organizations TNO, NFI, and SURF, in partnership with industry bodies like NDP Nieuwsmedia, whose members provided a substantial dataset. This project serves as a model for ethical AI development.
Also Read:
- OpenAI’s Atlas Browser Sidesteps Content from Suing Media Outlets
- Japanese Publishers Express Concerns Over OpenAI’s Sora 2 AI Tool and Copyright Infringement
Further emphasizing the industry’s stance, the NVJ (Dutch Association of Journalists) is actively advocating for a collective compensation scheme. This scheme aims to ensure that journalists and other creators receive fair remuneration when their work is utilized for AI training purposes. The ongoing debate highlights the need for clear legal frameworks and compensation mechanisms to address the challenges posed by AI’s reliance on vast amounts of online data, much of which is copyrighted.
