What happened

The Belastingdienst (Tax and Customs Administration) used an automated risk-scoring model to identify potential fraud in childcare benefit claims by flagging families for investigation based on discriminatory indicators, such as dual nationality or low income. 

Once flagged, authorities adopted a "zero-tolerance" approach, demanding the immediate repayment of years' worth of benefits, often tens of thousands of euros, over minor administrative errors like a missing signature. 

This led to thousands of families falling into deep debt, losing their homes, and being placed on secret blacklists. In over 2,000 cases, children were even taken into state foster care due to the resulting household instability. 

The Dutch government promised to reimburse each family EUR 30,000, only for the tax authority to demand much of it back in taxes. Eventually, the government cancelled the debts of those families impacted.

Welfare seekers were not informed about the system, and had little scope to complain or appeal decisions. Furthernore, external parties such as researchers, journalists or investigators were not permitted to inspect it.

However, the scheme's shortcomings and failures had been exposed many times, with examples of fraud against the system emerging soon after its launch in 2013, closely followed by cases involving suspected fraud, a number of which proved unfounded.

Substantive allegations of racial and ethnic profiling were subsequently levied by broadcaster RTL Nieuws and the Trouw newspaper in September 2018. Dual nationality and low income seen as major risk indicators, as was having a 'non-western appearance'.

The Toeslagenaffaire scandal led to the collapse of the Dutch government after a vote of no-confidence was called by opposition parties.

Why it happened

The root cause was a combination of discriminatory algorithmic design and a total lack of institutional accountability. 

The system was designed to penalise non-Dutch nationals, creating a "discriminatory loop" where minority groups were audited more frequently, which then "taught" the self-learning model that these groups were higher risks.

Furthermore, a "black-box" lack of transparency meant caseworkers were given risk scores without explanation, leading them to treat algorithmic suspicion as objective fact. 

A "perverse incentive" also existed: the tax authority was pressured to prove the efficiency of the new digital system by recouping high volumes of funds.