Instead of relying on large American IT companies, the Netherlands is introducing 'GPT-NL', its own public AI trained purely on legally obtained data, into everyday life, presenting a new model for European digital sovereignty.
Imagine this: you wake up in the morning, preparing for a move, and log onto your local city hall’s website to ask, “How do I get a home renovation permit in our neighborhood, and what are the waste sorting regulations?” Normally, you would have to spend ages digging through complicated civil service message boards and search bars, but now, a friendly chat window opens, instantly providing answers perfectly tailored to the local language and neighborhood laws.
But what’s surprising here is that this smart and friendly artificial intelligence (AI) assistant isn’t a system built by the giant American IT companies we all know. What if it were “our neighborhood’s dedicated AI,” transparently created within your own country using citizens’ tax dollars? And what if it were an assistant raised exclusively on perfectly clean and safe data, without a single copyright controversy?
Right now, in the Netherlands, this pleasant imagination is becoming a reality. The Netherlands is refusing the easy path of relying on the convenient systems of giant tech companies. Instead, they are setting a new milestone in European AI history by directly building their own unique language model, ‘GPT-NL’ [GPT‑NL: a sovereign language model for the Netherlands]. Why on earth did they decide to take the arduous route rather than use an AI like ChatGPT, which others have already spent astronomical amounts of money to build well? Hidden within this decision is a profound meaning that we all need to understand.
Why It Matters
Today, the voice assistants on the smartphones we use every day, translators, and large language models (LLMs, an AI technology that learns from vast amounts of text to grasp context and converse like a human) that summarize complex texts in an instant are mostly dominated by a few global American companies. Certainly, their technological prowess is astonishingly excellent and convenient. However, if you think a little deeper, there’s a dizzying aspect to it. Entrusting a country’s important public services, confidential tasks, and the core administrative infrastructure directly linked to the lives of ordinary citizens entirely to the servers and systems of foreign companies is a deeply unsettling prospect at the national level.
Amidst this anxiety, the most critical topic emerging globally in recent times is exactly ‘Digital Sovereignty’. Digital sovereignty means moving beyond simply importing and using the latest technology; it represents the independent right of a nation or its citizens to hold complete control over technology and data, unshaken by external forces. Simply put, it’s the difference between keeping the keys to our front door ourselves or entrusting them forever to a foreign security company across the ocean.
This analogy makes it much easier to understand: imagine that the sole source of drinking water in your village is monopolized by a massive multinational bottled water company located far across the sea. Right now, the water they provide tastes sweet, and it gushes out conveniently the moment you turn on the tap. But what if that company suddenly announces one day that they are raising the price of water tenfold, or if a mysterious problem in their filtration system causes them to pump out muddy water? The villagers would have no choice but to endure the thirst or bite the bullet and accept the unreasonable demands.
Therefore, the villagers pooling their money to dig their own well on their land and build their own transparently managed water filtration system—this is precisely the ultimate goal the Netherlands aims to achieve by investing immense effort and capital into developing GPT-NL [The Netherlands starts realisation GPT-NL..]. The Dutch government has made the courageous choice of ‘digital independence’ to reduce reliance on giant American tech systems and design a true European alternative to replace them themselves [Dutch project GPT-NL could change how Europe thinks about AI].
For this massive well-digging project, the Dutch government (through RVO under the Ministry of Economic Affairs) boldly invested a substantial public budget of 13.5 million euros (roughly 20 billion Korean won, a budget easily capable of building a state-of-the-art large library) [Netherlands moves GPT-NL from lab to live: first pilots under …]. And rather than commercial companies pursuing corporate profit, non-profit research institutions that prioritize the public good banded together. The Netherlands Organisation for Applied Scientific Research (TNO), the IT network organization for education and research (SURF), and the Netherlands Forensic Institute (NFI) joined forces to meticulously lay the foundation for this independent Dutch AI ecosystem from the ground up [GPT‑NL: a sovereign language model for the Netherlands]. Finally, in February 2025, they grandly weighed anchor, announcing the development of their own language model to the world [GPT-NL: Netherlands Builds Sovereign AI Language Model with …].
The Explainer
So, how does this GPT-NL created by the Netherlands differ technologically from existing famous conversational AIs? The most fundamental and decisive difference lies precisely in the purity of the ‘diet (data)’ that fills the AI’s brain.
The large-scale AIs we marvel at and use today must constantly gobble up massive amounts of text tokens (the smallest unit by which language models read and write text, much like Lego blocks or puzzle pieces) to become smarter. To achieve this, they often indiscriminately suck in and learn from millions of news articles, personal blog posts, pictures, and even painstakingly written novels floating around the global internet space. As a result, copyright infringement controversies naturally trail them like shadows for not seeking the original creators’ consent, often leading to a dark ‘black box’ phenomenon where even the developers do not know where or how the AI acquired its knowledge.
However, the Netherlands’ GPT-NL is completely different right from the starting line. The founder of this project proudly declared GPT-NL the “first legal-compliant language model” [Founder GPT-NL: ‘First legal-compliant language model’]. GPT-NL does not learn a single line of data scraped through dark channels or without the original author’s permission. The fact that it is the first large-scale Dutch AI model trained purely on so-called ‘clean data’, obtained legally and through a clear consent process, is their greatest weapon and point of pride [Dutch news publishers contribute to developing GPT-NL].
Put simply, if existing giant AIs are ‘omnivorous chefs’ who forcibly learned to cook by blindly scavenging everything from trash cans in restaurant back alleys to secret recipe notes hanging on someone else’s fence, GPT-NL is an ‘honest and transparent chef’ who officially paid a fair price to buy only fresh, safe ingredients directly from reliable local organic farms, solidly learning the basics of cooking from scratch.
In fact, to help birth this honest chef, representative media outlets (news publishers) in the Netherlands have rolled up their sleeves. To ensure the AI speaks proper and accurate Dutch and perfectly understands the local zeitgeist and culture, they are happily cooperating by providing their massive, legally owned collections of high-quality news articles as training data [[Large dataset news organizations for Dutch AI language model …](https://www.tno.nl/en/newsroom/2025/07/
References
- GPT‑NL: a sovereign language model for the Netherlands
- GPT‑NL: a sovereign language model for the Netherlands
- GPT-NL opens up: Inside the Dutch bid for sovereign AI
-
[Sovereign AI & Digital Autonomy: GPT-NL Model in Den… AetherLink](https://aetherlink.ai/en/blog/sovereign-ai-digital-autonomy-gpt-nl-model-in-den-haag) - Founder GPT-NL: ‘First legal-compliant language model’
- GPT-NL: An introduction to the Dutch language model
- The Netherlands starts realisation GPT-NL..
- Dutch project GPT-NL could change how Europe thinks about AI
- Large dataset news organizations for Dutch AI language model …
- Netherlands moves GPT-NL from lab to live: first pilots under …
- GPT-NL: Netherlands Builds Sovereign AI Language Model with …
- Dutch news publishers contribute to developing GPT-NL
-
[News Security Delta (HSD)](https://securitydelta.nl/news/overview/gpt-nl-a-sovereign-language-model-for-the-netherlands)
- It learned by randomly collecting all data on the internet.
- It is the first large-scale model trained exclusively on strictly legally obtained data.
- It was developed specifically for military defense purposes only.
- To fully secure national 'digital sovereignty' without relying on giant foreign IT companies
- To create the world's best all-purpose AI that solves math problems better than America's latest ChatGPT
- As part of a cultural policy to make Dutch the global lingua franca instead of English
- As a public administration assistant at The Hague City Hall helping guide municipal laws, permit requirements, and policies
- As a real-time automatic video translator for global social media platforms
- As the core navigation system for self-driving cars connecting all of Europe