Social benefits in the Netherlands

The child benefits scandal in the Netherlands

A highly publicised case from 2021 is the so-called child benefits scandal (“toeslagenaffaire”). For years, the Tax Administration used a fraud-detection algorithm that, based on patterns and indirect personal circumstances, mass-labelled families as high-risk.

What went wrong

The tool, intended to guide experts towards analysis and a more detailed review of a family’s situation, began to be used for decision-making about people’s entitlements. What was meant as support for oversight became, in practice, automated decision-making without expert judgement—resulting in unjustified sanctions and severe social consequences for tens of thousands of families.

Why the Netherlands is considered a reference point today

Because of the systemic changes that followed. They established a rule that an algorithm must not directly determine an individual’s entitlements; at the same time, they published a public register of algorithms used by the public administration, introduced impact assessments, and more.

Could AI be used for such a case today?

Yes—but only very thoughtfully and under strict oversight. A review of practice shows that AI-first systems are best introduced:

• From pilot solutions on real data to broader deployment. Today, pilot solutions do not test whether the technology can cover the use case; they test whether the technology is aligned with the use case. Sometimes you simply need to try—in a limited scope, with real data. And, if at all possible, with measurement.
• In teams that treat responsibility as an opportunity. Awareness of responsibility in the use of AI is crucial for the long-term success of the society we share. Once data leaks, you cannot get it back.

The AI Act and all related legislation may (still) not have reached a sufficient level of practicality and digestibility, but it has its purpose. Attention in this area is essential if we want to live well in the long run.