The tax office specifically targeted people with low incomes when checking for potential fraud involving childcare benefits, the agency has confirmed to RTL Nieuws and Trouw.

Between 2013 and July 2020, the tax office used a self-learning algorithm based on a risk classification system to decide who should face extra checks. The system was scrapped last year following a damning report but both RTL Nieuws and Trouw now say that income was one of the 20 or so criteria it used.

Asked by Trouw and RTL Nieuws about the income standard, the tax office said ‘there is a statistical link between income size and the chance of a wrongful application’.

Income, however, only formed a small part of the risk assessment and was included with the aim of preventing low income households having to pay back large amounts of money if their claim was wrongful, the tax office said.

Earlier it emerged that people without Dutch nationality, or dual nationals, were also seen as being more likely to commit fraud.

SP parliamentarian Renske Leijten has called on the finance ministry to publish the entire risk classification system and the government has promised to respond by the end of this week.

The system has also been slammed by the Dutch privacy watchdog AP and Amnesty International.

Last year, tax ministers Hans Vijlbrief and Alexandra van Huffelen told parliament that some 270,000 people had been flagged up as potential fraudsters, without any evidence or opportunity to defend themselves.