Inside Spain’s Secretive AI System for Detecting Sick Leave Fraud

Spain is known for its commitment to responsible AI within Europe. Yet, it has been discovered that the country’s transparency promises are often unmet, especially regarding algorithms used in justice and welfare. This lack of transparency is part of a broader trend where governments quietly use predictive algorithms to make significant decisions without public scrutiny. Independent investigations into such systems often reveal serious flaws and discriminatory practices.

Spain’s National Institute of Social Security (INSS) launched two machine learning algorithms in 2018 to identify fraudulent claims for sick leave benefits. These algorithms, meant to detect individuals potentially defrauding the state by exaggerating their illness, have operated in secrecy for years. However, public records and interviews reveal that these systems, while high-stakes, are opaque and unreliable.

The algorithms, developed by SAS and implemented by IBM’s subsidiary ViewNext, cost at least one million euros. Despite this investment, internal evaluations show the system generates numerous false positives and uses sensitive personal data, raising significant privacy concerns. Interviews with medical inspectors highlight the system’s inefficacy, suggesting that more staff, rather than flawed algorithms, would be a more effective solution.

This investigation, part of a four-part series with El Confidencial, uncovers the technical and ethical issues of the INSS’s AI system, comparing it to similar controversial practices in other countries.