EU AI Act Compliance Hurdles for Medical Risk AI Tools

A new preprint by Sami Andberg, Henri Terho, and Katja Saarela (University of Helsinki) examines the ethical and regulatory challenges of two medical AI tools: one predicting work disability risk (Case A) and another predicting Alzheimer’s disease risk (Case B). Using the EU AI Act as a framework, the researchers found both tools meet the criteria for high-risk AI systems—meaning they must undergo rigorous conformity assessments, including risk management, data governance, transparency, and human oversight. The paper, presented at the 2026 International Conference of Breakthroughs in AI (BAI’26) and published in Lecture Notes in Networks and Systems, notes that while the tools show promise for early intervention, transitioning from research to production demands significant compliance work.

The analysis underscores a growing tension between AI’s potential in healthcare and the regulatory burden intended to protect patients. For the work disability predictor, ethical concerns include potential bias against certain demographics and the psychological impact of knowing one’s risk. For the Alzheimer’s predictor, issues of informed consent and the lack of effective treatments raise unique ethical dilemmas. The authors argue that developers must integrate ethical AI principles—fairness, accountability, and transparency—from the design phase to satisfy both regulators and society. The study serves as a roadmap for how medical AI can navigate the EU AI Act, but also warns that current compliance costs may deter smaller innovators.