Udbetaling Danmark's Welfare Fraud-Control Algorithms — Amnesty Finds Up to 60 Models and Risks of Discrimination, the Agency Disputes It

Udbetaling Danmark (UDK), the Danish public benefits payment agency, and ATP, the statutory pension and social-security body that hosts it, use a fraud-control system that Amnesty International says relies on up to 60 algorithmic models to flag people for investigation. In its report 'Coded Injustice: Surveillance and Discrimination in Denmark's Automated Welfare State' (13 November 2024), Amnesty reports that inputs include residency status and movements, citizenship, place of birth and family relationships, which it argues can act as proxies for race, ethnicity or sexual orientation.
Two named models illustrate the concern: 'Really Single' predicts a person's relationship status, and 'Model Abroad' prioritises people with medium or high-strength ties to non-EEA countries for investigation. Amnesty concludes that the system risks discrimination against people with disabilities, low-income people, migrants, refugees and marginalised racial groups, and functions as a social-scoring system that it argues should be prohibited under the EU AI Act.
UDK disputes this: it says the data collection is legally grounded, that using citizenship as a parameter does not amount to processing sensitive data, and that its approach is unlikely to fall under the AI Act's social-scoring ban. Amnesty states UDK and ATP supplied only redacted design documents and rejected requests for a collaborative audit.
This is a cautionary case: the findings come from a civil-society investigation rather than an independent audit, no published impact evaluation shows measured public value, and the legal question remains contested. It is valuable for the transparency and fairness lessons it raises for any welfare fraud-scoring system.

Read the full analysis: https://www.amnesty.org.uk/latest/denmark-new-report-mass-surveillance-and-discrimination-automated-welfare-state