Försäkringskassan's Fraud-Risk Algorithm — Sweden's Decade-Old Welfare Model Confronts Its Own Bias Evidence

Försäkringskassan, Sweden's Social Insurance Agency, used a machine-learning model from 2013 to assign fraud-risk scores to social-security applicants and decide which cases to flag for investigation.
A November 2024 investigation by Lighthouse Reports and Svenska Dagbladet, analysing a dataset of more than 6,000 people flagged by the algorithm in 2017, found the model systematically 'overselected' women, applicants with foreign backgrounds, low-income earners and people without university degrees for investigation — while performing poorly at identifying men and higher earners who had actually committed fraud; very few flagged cases led to a court finding of intentional fraud.
Sweden's own inspectorate, ISF (Inspektionen för socialförsäkringen), first raised equal-treatment concerns about the model in 2018. Following the 2024 media reports, the data-protection authority IMY opened a formal supervision case in June 2025 into whether the model's use of personal data complied with GDPR.
In its response to IMY on 5 September 2025, Försäkringskassan confirmed it had discontinued the machine-learning risk profile for parental-benefit checks about a month earlier and had no plans to reactivate it. IMY closed the case on 18 November 2025, concluding that since the system was no longer in use, the risks it examined had ceased — making this the most concrete resolution of the three Nordic/Balkan welfare-algorithm cases in this batch: full discontinuation confirmed by an independent regulator, rather than continued use with added oversight.

Read the full analysis: https://www.riksdagen.se/sv/dokument-och-lagar/dokument/skriftlig-fraga/forsakringskassans-anvandning-av-ai_hc11728/