evidoria

← Back to browse

Good practice Imported

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

Sweden · Stockholm · See the Sweden profile · See the Stockholm profile

Evidence: Observational / pre–post Top 32% 67/100 · Ask Evidence Copilot about this practice

Sweden's Social Insurance Agency scored applicants' fraud risk by algorithm from 2013. A 2024 Lighthouse Reports probe found it overselected women and low-income/foreign-background applicants; the agency retired the model in 2025 and IMY closed its case that November.

>6,000 people flagged by the algorithm in 2017
Dataset analysed by investigators (2017 data, analysed Nov 2024)
2013-2025 (~12 years)
Years in production before discontinuation (2013-2025)
Case closed 18 November 2025 following confirmed discontinuation
IMY GDPR supervision case outcome (Jun-Nov 2025)
Försäkringskassan's Fraud-Risk Algorithm — Sweden's Decade-Old Welfare Model Confronts Its Own Bias Evidence

Details

Maturity
Discontinued
Promoter
Försäkringskassan (Swedish Social Insurance Agency)
Period
2013–2025
Region (NUTS)
SE11
Keywords
social insurance, fraud detection, algorithmic risk scoring

Context

Sweden's Social Insurance Agency, Försäkringskassan, used a machine-learning model from 2013 to assign fraud-risk scores to social-security applicants and decide which cases to investigate.

Objectives

Score applicants for fraud risk to target investigation resources more efficiently.

Activities

The model ran in production from 2013. Sweden's inspectorate ISF first raised equal-treatment concerns in 2018. In November 2024, Lighthouse Reports and Svenska Dagbladet analysed a dataset of more than 6,000 people flagged by the algorithm in 2017. Following the reporting, the data-protection authority IMY opened a formal GDPR supervision case in June 2025.

Results

The 2024 investigation 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. Försäkringskassan confirmed to IMY on 5 September 2025 that it had discontinued the model about a month earlier with no plans to reactivate it, and IMY closed its case on 18 November 2025.

Conclusions

Independent journalistic and regulatory scrutiny documented clear demographic bias in a decade-old welfare fraud-risk algorithm, leading to full discontinuation confirmed by both the agency and the data-protection regulator.

Implementation
Indicative cost
Medium (€50k–€500k) — A machine-learning risk-scoring system operated in production across a national welfare agency's applicant base for over a decade, implying a medium ongoing operating cost before discontinuation.
Time to results
Long (> 3 years) — In use from 2013 to its discontinuation around August 2025 (confirmed to the regulator 5 September 2025), with the regulatory case formally closed 18 November 2025 -- a 12-year lifecycle.
Staffing & skills
Försäkringskassan (Swedish Social Insurance Agency) fraud-investigation and IT staff, ISF (Inspektionen för socialförsäkringen), national inspectorate, IMY (Swedish data-protection authority)

Conditions for success

  • Independent inspectorate and regulator oversight able to investigate and act on bias concerns
  • Willingness to discontinue a system once bias is confirmed, and to confirm this in writing to the regulator

Common failure modes

  • The model systematically overselected women, low-income and foreign-background applicants while missing actual higher-earning male fraud cases, for over a decade before correction
  • It took a media investigation, not routine internal governance, to surface the bias despite the inspectorate ISF having raised equal-treatment concerns as early as 2018

Where it fits

Governance type
national social insurance agency
Scale
nationwide applicant screening
Income level
high-income

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful