FDA official says AI medical tools should be tested like doctors
An FDA digital health director proposes that generative‑AI medical tools be evaluated with the same rigor as physicians.
21 Oct 2026 0 reads Compiled with AI
#FDA #Generative AI #Medical devices #Radiologist #AdvaMed #Digital Health
The MedTech Conference in Boston brought together regulators and industry leaders, and at the event FDA’s Digital Health Center of Excellence director, radiologist Rick Abramson, highlighted a shift in how artificial intelligence might be regulated. Abramson, a Harvard‑trained radiologist, explained that generative‑AI tools could be assessed in a way that mirrors the evaluation of a clinician’s competence rather than the traditional software‑as‑a‑medical‑device model.
How AI differs from traditional software
Current FDA oversight for medical software relies on a straightforward input‑output paradigm: a piece of software receives a defined signal, such as a chest X‑ray, and produces a clear result, like a binary yes or no for pneumothorax. This deterministic approach allows regulators to test safety and efficacy by running the software on a finite set of cases, a process the FDA has used for the past two decades.
Generative AI, however, changes the equation. Its output is not a fixed binary; it can produce limitless text or images in response to any input, creating an “unbounded evaluation space.” Abramson noted that this variability means the existing evidentiary methods are inadequate, and that AI tools might need to be judged on clinical judgment, communication, and adherence to safety boundaries much like a physician.
To address this gap, the FDA released a discussion paper in August that invited industry feedback on how to regulate such tools. The comment period ended with more than 300 responses, indicating strong interest from stakeholders. Abramson said the agency intends to move beyond a scholarly exercise and draft concrete policy that reflects this input.
Looking ahead, Abramson indicated that a draft guidance could appear within months, after a careful review of the comments. He emphasized that accountability would remain with clinicians, patients, institutions, and developers, and that the policy process would involve further stakeholder engagement before finalization.