Regulatory & Compliance
The EU AI Act and clinical decision support
How the EU AI Act and MDR classify clinical decision-support AI as high-risk, and the staged path to Class IIa certification Elna is building toward.
Pau Rue
Founder at Elna
The EU AI Act sorts AI systems by risk, and puts most clinical decision-support software into its highest regulated tier: high-risk. If an AI tool is used to inform a diagnosis or a treatment decision, the Act, together with the EU Medical Device Regulation (MDR), expects it to meet the same bar as other medical software: a quality management system, documented risk management, clinical evidence, and ongoing post-market monitoring.
That’s a real bar, and it’s worth understanding rather than treating as fine print.
Two categories, one regulatory line. Under MDR, software that retrieves and presents published medical literature, without generating a patient-specific recommendation, is not classified as a medical device. The moment a tool moves from “here’s what the evidence says” to “here’s what you should do for this patient,” it crosses into Software as a Medical Device (SaMD) territory, and under MDR Annex VIII Rule 11, decision-support software of that kind is classified Class IIa by default (escalating higher only where errors could cause severe or irreversible harm).
Where Elna stands today. We currently operate under a non-device positioning: a medical information retrieval system for healthcare professionals, with guardrails that keep it there, no patient-identifying inputs, output filtering against patient-specific formulations, audit logs. That’s a deliberate, legally scoped Phase 1, not a way of avoiding the question.
Where we’re headed. Our Phase 2 target is full MDR Class IIa certification, keeping a licensed clinician as the decision-maker throughout (human-in-the-loop) rather than pursuing autonomous decision-making. This isn’t an open regulatory question we’re hoping resolves in our favor, it’s a proven path. Two European generative-AI clinical tools have already cleared it: Tonic App, certified Class IIa by BSI, and Tandem Health, whose clinical decision support went live under a Class IIa CE mark in 2026.
Why does this matter to a hospital evaluating clinical AI today? Because “we’ll figure out compliance later” isn’t a viable position once a tool touches patient-specific decisions, and because the AI Act’s high-risk obligations and MDR’s device requirements now move together. A vendor who can show you a staged, costed, precedented path through both is telling you something concrete about how seriously they take the second half of “safe and effective.”