The U.S. Food and Drug Administration’s (FDA) recently released discussion paper on generative artificial intelligence (GenAI)-enabled medical devices may be as notable for what it does not say as what it does say. The Considerations for the Regulation of Generative AI-Enabled Medical Devices: Discussion Paper and Request for Feedback is for discussion purposes only; it is not agency guidance. Interspersed through the paper are 26 discussion questions, on which FDA is encouraging feedback by October 19, 2026.
The discussion paper focuses on the “transformative promise” but “unique risks” of medical devices enabled by GenAI. While FDA has previously recognized this dynamic for artificial intelligence (AI) generally, this discussion paper omits much of the nomenclature FDA (and to some extent, Congress) has previously used in evaluating these issues. For example, the discussion paper does not acknowledge FDA’s previous distinction between “locked” and “adaptive” algorithms or discuss how the use of GenAI fits within this previously emphasized distinction. The discussion paper also acknowledges that a GenAI-enabled software function can be “integrated into a broader device,” but assiduously avoids referring to the previously preferred terminology of software in a medical device (SiMD) and software as a medical device (SaMD). Finally, while the paper recognizes the importance of how “independently” a software function “directs or takes action,” it omits reference to the well-trod concept of “clinical decision support” (other than in a footnote). While FDA may be burying or phasing out these concepts, the broader principles and considerations are familiar, albeit with some different packaging.
FDA reiterates the general principles of a risk-based approach and application of least burdensome principles. The discussion paper also identifies previously expressed considerations for the appropriate evaluation and regulation of GenAI-enabled devices, such as the challenge of reviewing such devices solely in the premarket setting, and the potential need to complement pre-market evidence with post-market evidence. The discussion paper also offers up a two-axis framework for evaluating risk for GenAI-enabled devices. While not exactly the same as the two-axis concepts proposed for evaluating software functions in the past, the general considerations echo the past considerations, as FDA identifies the consequences of an incorrect output as one key factor and the activity being performed (and its level of independence) as the other. In addition, the discussion paper recognizes that patient-facing functions “may present different considerations” than health care professional (HCP)-facing functions—a distinction that FDA has considered regarding software-enabled devices for decades.
The discussion paper indicates that FDA is considering a “competency-based approach” to the evaluation of GenAI-enabled devices. FDA describes a model “inspired by the evaluation and credentialing of human clinicians,” noting that “clinicians are not evaluated through exhaustive testing of every scenario they may encounter; instead, they are evaluated through a combination of structured assessments of underlying knowledge and reasoning….” The discussion paper goes on to describe how a similar approach could be applied to GenAI-enabled devices.
Ultimately, the discussion paper describes a potential regulatory approach that, while sharing many common objectives and considerations with the current risk-based device framework, would require dramatic changes to existing regulatory and review paradigms. It would almost certainly require statutory changes as well. The 26 discussion questions wade into many of the details of how such a framework would, and should, operate. Providing detailed and thoughtful input on all of these questions by the October 19th deadline will be a significant undertaking but provides an important opportunity for stakeholders.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
[View Source]
