SwedenPublic sector2013 – ongoingFörsäkringskassan (Swedish Social Insurance Agency)
Machine-learning risk scores for benefits fraud control
Investigation finds Sweden's benefits fraud scoring flags women and people of foreign background more often (Nov 2024)
Problem
Försäkringskassan has used a machine-learning system since at least 2013 to score benefit applicants. Those with the highest scores are sent to the fraud-control unit for investigation.
Approach
Lighthouse Reports and Svenska Dagbladet tested the system against six fairness metrics and found it flagged women, people with foreign backgrounds, low earners and people without university degrees disproportionately. The agency had refused freedom-of-information requests and argued that a 2018 oversight analysis rested on dubious grounds; Amnesty asked for the system to be stopped.
Results
No figures published
Takeaway
Journalists found that the system disproportionately selected several groups for fraud investigation, prompting Amnesty to call for it to stop under the EU AI Act's duty to assess rights risks before deployment. The agency disputed a 2018 oversight analysis.
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
- Built in-house
- Evidence
- Press reporting
Sources
Source summaries
Amnesty reports the findings of Lighthouse Reports and Svenska Dagbladet on the agency's risk scoring. It identifies the groups flagged more often and says the agency refused information requests.
Full details are in the original sources.
