How AI is changing work
Government
Working faster, with explanation and oversight.
Municipalities, implementing agencies and Rijkswaterstaat increasingly use AI, mainly for internal work such as summarising, anonymising and forecasting. Government has also learned what goes wrong when algorithms decide without oversight. That is why transparency and a human who decides belong with every application.
What changes
- Summarising documents, transcribing and drafting policy texts are done with AI assistants.
- Documents for Woo (freedom of information) requests are anonymised with AI, with an employee who checks.
- Models for water, maintenance and planning become faster and more precise.
- Algorithms are publicly described in the Algoritmeregister (algorithm register).
Where people make the difference
- Making and explaining decisions about people, with an eye for the consequences.
- Testing whether an algorithm works out fairly for different groups.
- Contact with residents when things get complicated or personal.
How AI is already used
Anonymising without black bars
The municipality of Hoeksche Waard, together with five other municipalities, has a language model replace names in documents with neutral terms, so documents stay readable.
Computable, 2026Suggestions, not decisions
Zaanstad uses an AI tool that suggests what should be redacted in Woo documents. Staff accept or reject each suggestion; the tool is described in the Algoritmeregister.
AlgoritmeregisterWater models with AI
Rijkswaterstaat, Deltares, KNMI and DigiShape are researching how AI can make water models faster and more flexible, for forecasts of high water, drought and shipping.
Rijkswaterstaat, 2025An assistant of its own for the judiciary
Staff of the Rechtspraak, the Dutch judiciary, write, improve and summarise memos and policy documents with RechtspraakGPT. It is explicitly not intended for judicial decisions.
Mr. OnlinePseudonymising rulings and planning
The Rechtspraak wants to use AI for tasks such as pseudonymising rulings, planning and summarising large case files, initially only where the risk is low.
AdvocatenbladAgents at the Tax and Customs Administration
An exploratory study advises the Belastingdienst, the Dutch Tax and Customs Administration, to start with document analysis, searching regulations and standard questions.
iBestuurHandling objections faster
Amsterdam tested transcription and summaries of formal objections; Breda tested an assistant for documents and analyses.
Gemeente.nu
Where the sector is heading
The number of generative AI applications in government grew from 8 in 2024 to 81 in 2025. 37 of them (46 percent) are already part of everyday work.
TNO, 2025On 5 October 2026, the Algoritmeregister contained 1,574 algorithm descriptions from 354 publishing organisations.
AlgoritmeregisterFor years, the Dutch Tax and Customs Administration processed the (dual) nationality of childcare benefit applicants in a discriminatory way, including as a risk indicator. In 2021, the AP, the Dutch data protection authority, imposed a fine of 2.75 million euros.
Autoriteit Persoonsgegevens, 2021In 2024, the Dutch cabinet apologised for indirect discrimination in DUO's checks on the student grant for living away from home: students from neighbourhoods with many residents with a migration background were checked more often.
Rijksoverheid, 2024
What a maker can do for you here
- “Build an assistant that summarises formal objections for our lawyers.”
- “Set up a safe AI environment for staff, with agreements about personal data.”
- “Make explainer videos about our schemes in plain language.”
- “Help us describe and test an algorithm for the Algoritmeregister.”
Researched on 5 October 2026. Know a good example from the Netherlands? Let us know.