The EU AI Act treats AI used in medical contexts as high-risk. For a vendor like Faurenis that means several concrete obligations: a quality-management system, technical documentation, transparency to clinicians, human oversight, and accuracy and robustness logging.
We are reading the Act conservatively. Two practical consequences:
First, every AI surface in Faurenis must support human oversight in a way the clinician can actually exercise. That is more than a "review" button — it is the badge-every-AI-paragraph design choice from our earlier post, plus the validation queue, plus the transcript-line grounding. These exist in part because the clinician needs to be able to override at the right granularity, not just at the end.
Second, accuracy logging is a product feature, not just a compliance one. We log per-model, per-task accuracy on a held-out evaluation set and report it inside the admin console. If accuracy drops, the admin sees it before a clinician feels it.
We will not claim "EU AI Act compliant" until the regulatory landscape and certification pathway are fully formed. We will say "designed for the EU AI Act" — which is what the work in the product currently is.