FDA is considering a competency-based approach to evaluating generative AI-enabled medical devices, similar to assessments for doctors, according to Axios.

A competency-based framework, testing AI devices the way clinicians are examined, would be a departure from the FDA's established device paradigm, which evaluates a fixed product against safety and efficacy endpoints at a point in time. The agency has wrestled for years with how to regulate adaptive software, and prior attempts have centred on predetermined change control plans: pre-authorising bounded model updates rather than re-reviewing each iteration. A competency model would shift the burden toward ongoing performance assessment, which cuts both ways for the peer set of listed device and diagnostics names: faster iteration for approved products, but a moving post-market compliance obligation in place of a one-off clearance. Worth noting this is a consideration stage report via a single media outlet, not draft guidance; historically, FDA thinking at this stage often takes multiple quarters to surface as formal documentation, and the final framework frequently differs materially from early signals. The tells are whether the agency's digital health centre publishes discussion papers, whether device trade bodies respond formally, and whether any pilot programme with named manufacturers emerges. Read-through to AI-exposed healthcare names is real but second order at this stage.