In 2021, the Dutch government collapsed following revelations that thousands of families had been wrongly accused of childcare benefits fraud. At the heart of the scandal lay an algorithmic risk-profiling system that disproportionately targeted vulnerable and migrant families, labelling them as “high risk” based on opaque data correlations. As European states integrate artificial intelligence (AI) into public administration, a pressing human rights question emerges: can algorithmic risk assessments truly serve the best interests of the child?

Across Europe, public authorities are exploring data-driven tools to support decisions in child protection and youth justice systems. These systems promise consistency and efficiency, identifying patterns of potential neglect or harm through statistical modelling. Yet predictive technologies operate on historical datasets. Where those datasets reflect socio-economic inequalities, over-surveillance of certain communities, or institutional bias, algorithmic outputs may reproduce and amplify existing disparities rather than mitigate them. Although the Dutch childcare benefits scandal did not arise directly within child protection law, it exposed a structural danger increasingly relevant across Europe: when algorithmic systems classify families through predictive models, the consequences may profoundly affect children’s rights and family integrity.

This tension strikes at the core of Article 3 of the UN Convention on the Rights of the Child (CRC), which requires that the best interests of the child be a primary consideration in all actions concerning children. As clarified by the UN Committee on the Rights of the Child in its General Comment No. 14 (2013), this principle requires an individualised assessment grounded in the specific circumstances of each child, taking into account their unique situation, needs, and vulnerabilities.

Similarly, decisions that interfere with family life engage Article 8 of the European Convention on Human Rights (ECHR), and are only justified where legal, necessary, and proportionate to the legitimate aims pursued by the interference. When algorithmic risk scores significantly influence decisions about monitoring, intervention, or potential removal, transparency and individualised judicial oversight become essential safeguards.

The Dutch childcare benefits scandal illustrates how reliance on automated risk categorisation can undermine procedural fairness. Affected families often struggled to understand how decisions were reached or to effectively challenge risk classifications, as documented in official investigations and public reporting on the scandal. In child protection contexts, similar opacity could weaken trust in institutions and compromise the legitimacy of state intervention.

This does not mean that technology has no role in safeguarding children. Data analysis may help identify systemic gaps or support early intervention strategies. As illustrated by the Dutch childcare benefits scandal, automated risk assessments may flag families as “high risk” without adequately accounting for individual circumstances, leading to disproportionate and unjust outcomes. However, algorithmic tools must remain subordinate to human rights principles. The best interests of the child require more than statistical probability; they demand contextual sensitivity, proportionality, and meaningful judicial oversight.

The broader European regulatory landscape reinforces this concern. The emerging framework of the European Union Artificial Intelligence Act recognises that certain AI systems used in public administration and law enforcement pose “high-risk” implications for fundamental rights. Where such systems affect children and families, heightened scrutiny is not optional but necessary. Algorithmic efficiency cannot replace individualised human judgment grounded in dignity and fairness.

As Europe moves toward greater reliance on AI-driven governance, child protection and youth justice systems must resist technocratic overconfidence. Algorithms may inform decisions but they cannot determine them. A rights-based approach requires that predictive tools operate within clear normative boundaries, ensuring that the best interests of the child remain a living legal principle rather than a statistical abstraction. If algorithmic governance expands without robust, rights-based safeguards, the promise of child protection may quietly transform into a new architecture of digital suspicion.