Much greater openness and collaboration within government on GenAI

Less reluctance

‘Two years ago, there was still a reluctance around GenAI to share information with one another or with the outside world,’ say AI experts Marissa Hoekstra and Tessa Bruijne of TNO Vector. Commissioned by the Dutch AI Coalition, TNO Vector carried out a quick scan two years ago into the use of AI applications within government, followed at the end of last year by an in-depth study specifically focused on generative AI. The intention is to repeat this monitor annually.

From caution to encouragement

In the 2024 quick scan, TNO identified 266 AI applications within government, of which only eight involved GenAI. This has now increased to 81. In 2024, GenAI was still in its infancy and government bodies were cautious about experimenting with and deploying it, whereas today this is actively encouraged. Promoting transparency and communication, which were explicitly identified as points of attention in the quick scan, now appears to be standard practice.

'Government organisations are increasingly turning to one another to share experiences with AI. AI is being adopted in different ways, from in-house development to procurement and collaborative partnerships.'

^{AI experts Marissa Hoekstra and Tessa Bruijne of TNO Vector}

Extensive collaboration

‘Government organisations are increasingly approaching one another with questions such as: how did you tackle this, what are your experiences, what should you pay close attention to? Some organisations choose to develop AI in-house, while others opt to procure it. There is also a great deal of collaboration. In other words, there are multiple ways in which AI is currently being adopted within organisations.’

Practical examples available

TNO Vector conducted research at ministries, municipalities, provinces and agencies such as RVO, the National Archives and the KNMI. Interviews were held to understand how GenAI is developing within these organisations. The findings have now been made publicly available. The Platform AI and Government website features dozens of practical examples, each with a brief description of the GenAI application and the parties involved. Using filters such as status, type of organisation, domain or application, professionals working with AI and other interested parties can explore these examples in greater depth.

Primarily textual applications

‘The quick scans from 2019, 2021 and 2024 provided insight into how AI is developing within government organisations. In 2025, we focused specifically on GenAI and the success factors and bottlenecks encountered in practice. We see that GenAI and language models within government are primarily used to support textual applications. Government holds vast amounts of textual documentation, and GenAI helps to search, summarise, translate and clarify this information far more quickly. Initially, this is mainly for internal use, but increasingly applications are also being developed for citizens and businesses.’

'We are increasingly hearing from government organisations that their existing IT software suppliers are suddenly offering AI functionality as well.'

^{AI experts Marissa Hoekstra and Tessa Bruijne of TNO Vector}

Low threshold

In a short period of time, GenAI has also matured and become accessible to a broad audience. There are free AI tools available online, as well as many off-the-shelf solutions that can be readily procured, with a wide choice of suppliers. ‘This has made the adoption of AI applications within organisations very accessible.’

At the same time, existing IT system providers are also integrating AI functionalities into their software. ‘We increasingly hear from government organisations that existing IT vendors are now suddenly offering AI functionality. As an organisation, you need to consider carefully whether or not you want this incorporated into your systems.’

Preventing shadow use among employees

One of the bottlenecks identified by the researchers concerns privacy risks. Organisations observe that employees are keen to use free AI tools such as ChatGPT. However, these tools carry significant risks. For example, their use may lead to data breaches if sensitive (personal) information is entered into them. To prevent such ‘shadow use’ among staff, many government organisations aim to provide a safe and responsible alternative.

‘Many governments are therefore looking for alternative models trained on internal data and equipped with the appropriate security measures. Numerous pilots are already underway in this area.’

Alternatives to US-based language models

Another bottleneck is digital dependency. Many GenAI applications rely on external providers for models and infrastructure and are based on US-developed models. This raises questions about security and digital sovereignty. Government organisations are actively exploring how they can become less dependent on US suppliers. TNO is also working with SURF and the NFI on GPT-NL, a Dutch language model aligned with our values and standards.’

Strengthening competencies

Within central government, as well as the Association of Netherlands Municipalities (VNG), various initiatives are underway to ensure that AI applications developed in one place become more widely available to other government organisations. TNO Vector is also developing instruments to stimulate the scaling up of AI applications and is exploring how AI competencies and literacy can be strengthened across government.

'AI has enormous potential to help government address societal challenges, provided it is deployed responsibly.'

^{AI experts Marissa Hoekstra and Tessa Bruijne of TNO Vector.}

AI encompasses more than GenAI

In addition to GenAI, organisations are increasingly making use of other AI techniques. ‘While GenAI within government is still largely limited to textual applications, the 2024 quick scan already showed that government organisations are also experimenting with other AI techniques, such as machine learning and image recognition. For example, in the physical environment, drones equipped with image recognition technologies are used for monitoring and sensing, such as inspecting bridges or detecting fishing traps. In various policy domains, experiments are underway with machine learning techniques for predictive purposes to support policymaking. AI has enormous potential to help government in many areas and to address societal challenges, provided it is used responsibly. GenAI represents only a small part of what AI is or can be. Our call is therefore for organisations not to focus solely on GenAI, but also to remain mindful of the opportunities and risks of other forms of AI,’ the researchers conclude.