AI-verktyg för journalföring - Klassificering och rättsliga konsekvenser enligt AI-förordningen

Sammanfattning

In recent years, artificial intelligence has gained increasing importance within the healthcare sector. One emerging application is AI medical scribes, which automate parts of administrative work by converting speech to text, structuring clinical notes, and in some cases suggesting diagnoses or referrals. These tools have the potential to improve the efficiency of healthcare professionals, but at the same time raise complex legal questions. With the new EU Artificial Intelligence Act (2024/1689) (the AI Act), these systems will be classified according to their level of risk, and this assessment will have major implications for both providers and deployers of AI tools, including those used in healthcare. The aim of this thesis is to examine how AI medical scribes should be classified under the AI Act and what legal consequences this entails for providers of these tools and healthcare providers that use them. To answer this question, an EU legal dogmatic method is applied, supplemented by elements of legal analysis. The AI Act divides AI systems into four risk categories (unacceptable, high, limited, and low risk) with regulatory requirements increasing significantly with the risk level. The thesis shows that for AI systems in healthcare, the decisive boundary lies between high and low risk. AI medical scribes that qualify as medical devices under the MDR are automatically classified as high-risk. A high-risk classification entails extensive requirements regarding documentation, risk management, data governance, human oversight, and CE marking, whereas low-risk systems are essentially left unregulated. This substantial difference in regulation has major practical and economic implications. The classification issue is however unclear for AI medical scribes that do not qualify as medical devices and operate in the grey zone between administrative and clinical support. The ambiguity arises because a high-risk classification extends beyond MDR products to also include AI systems intended to evaluate eligibility for healthcare services. However, certain simpler systems, such as those performing only narrow and procedural tasks, may be exempted, thus complicating the classification boundary. The analysis shows that simpler medical scribes that merely perform narrow procedural tasks, such as transcription, are likely to be exempt and classified as low risk, while more advanced systems that influence clinical decision making are likely to be considered high risk. Where the boundary lies remains uncertain, which is problematic given the substantial differences in regulatory obligations. This uncertainty, in turn, risks discouraging innovation and technological development within healthcare. In conclusion, the thesis finds that the AI Act establishes a necessary structure for safeguarding patient safety and strengthening trust in AI within healthcare, but that the current legal situation remains marked by significant uncertainty. A key future challenge will be to develop clearer guidance, harmonised standards, and a more predictable application of the AI Act within the healthcare sector, allowing patient safety and innovation to be promoted in parallel.