The US Food and Drug Administration is seeking public feedback on how generative AI features embedded in medical devices should be regulated, opening a consultation that will shape the pathway for clinical software whose behaviour changes after authorisation. The question is technically awkward because medical device clearance has historically been premised on a fixed, validated function, whereas a generative model's outputs vary between deployments and can shift with updates to the underlying model — the same property that has already driven the FDA's work on predetermined change control plans for adaptive algorithms. For Australian and New Zealand providers the consultation matters through equivalence: the TGA and Medsafe both lean on FDA and EU precedent when setting expectations for software as a medical device, so the shape of the American framework tends to arrive locally within a procurement cycle. It also lands in a week where healthcare's AI story has been dominated by integration risk rather than model risk — the Mirth Connect flaws reported yesterday and the Orthanc advisory above — a reminder that clinical AI enters through the same unglamorous middleware as everything else in a hospital.

• Source article