Finland’s social security agency abandons AI trial for fraud detection over legal uncertainty

Finland’s Social Insurance Institution (Kela) has halted a pilot project using artificial intelligence to detect welfare fraud due to unclear legal boundaries, Yle reports.

The agency had tested an AI system to flag suspicious anomalies in benefit claims but abandoned the trial last autumn after determining that Finnish law does not clearly permit such use. Nina Nissilä, Kela’s director of IT services, cited insufficient legal precedent, stating that interpretations of AI applications in this context remain ambiguous.

“Current case law is incomplete, and there is no unified understanding of where the acceptable limits of these algorithms lie,” Nissilä said.

The AI was designed to scan large datasets for irregularities, which would then be reviewed by human officials for potential follow-up. However, Kela opted not to risk being the first to test legal boundaries. A separate AI tool for detecting document forgery was also discontinued, though Kela has since adopted a non-AI solution for this purpose.

In 2023, Kela identified €7.2 million in welfare fraud—less than 0.05% of its total €16.5 billion in disbursed benefits—with most cases involving basic social assistance. While the number of fraud incidents has declined, the average sum per case has risen, keeping the total figure stable. Kela filed 471 police reports out of 1,017 suspected fraud cases last year.

The decision follows international controversies over AI-driven welfare surveillance. In Sweden, the equivalent agency’s algorithm was found to disproportionately target marginalised groups based on gender, ethnicity, income, and education level, prompting its discontinuation. Human rights group Amnesty International has raised concerns about similar risks in Finland’s case.

Kela has not ruled out revisiting AI fraud detection once its €589 million Eepos IT modernisation project—partially AI-reliant—advances. Nissilä suggested clearer national guidelines could enable future reassessment, noting that insurers and other sectors face similar dilemmas.

Jarkko Levasma, Finland’s government CIO, acknowledged the legal ambiguity, stating that current laws do not definitively separate permitted from prohibited AI uses in public administration. While fully automated decisions by AI are banned, partial applications remain a grey area.