DenmarkPublic sector2024 – ongoingUdbetaling Danmark (UDK) and ATP
Algorithmic fraud-control models for social benefits
Amnesty says Denmark's benefits authority runs up to 60 fraud-detection models that risk discrimination (Nov 2024)
Problem
UDK and the pension body ATP use up to 60 algorithmic models to flag people for fraud investigation. Amnesty got partial access to four of them.
Approach
Amnesty says the models, built with the company NNIT, pool data on millions of residents and can discriminate against disabled, low-income and migrant people. UDK rejected Amnesty's view that the system amounts to social scoring and turned down a joint audit with full access to code and data.
Results
Takeaway
UDK declined a joint audit with full access to the code and data used to flag benefit recipients for fraud investigation. Amnesty's discrimination concerns rested on partial access to four models, and the agency disputed them.
Categories
- Size
- Large enterprise
- Industry
- Public sector
- Business function
- Finance, risk and legalProduction and quality
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Co-developed with a partner
- Evidence
- Independent study
Sources
Source summaries
Amnesty investigated Denmark's use of up to 60 fraud-control models, including 'Really Single' and 'Model Abroad'. UDK rejected the social-scoring charge and declined a joint audit.
Full details are in the original sources.
