FDA Weighs 'Competency-Based' Review for AI Medical Devices
The agency is considering a shift toward evaluating generative AI tools using benchmarks similar to those used for human clinicians.
The U.S. Food and Drug Administration is considering a new framework for evaluating generative AI-enabled medical devices that mimics the way human doctors are credentialed. This proposed shift would move the agency away from purely static performance metrics toward a competency-based approach to validate clinical reasoning.
According to a report from Axios, the FDA is exploring a model that sets specific benchmarks to confirm how AI tools perform in actual clinical settings. This process could involve comparing the AI's output against a panel of qualified clinicians or the performance of a median clinician to ensure the tool exhibits sound judgment. This initiative comes as the volume of AI integration in healthcare accelerates; the FDA has authorized over 1,000 AI-enabled medical devices to date, with industry reports indicating that between 258 and 295 of those were authorized in 2025 alone.
The Regulatory Challenge
Historically, the FDA has relied on pathways such as the 510(k) or De Novo processes, which typically require a device to demonstrate "substantial equivalence" to a product already on the market. While effective for hardware, these static approvals are often insufficient for machine learning models that are dynamic and can evolve after deployment. Consequently, there is a growing industry push for Total Product Lifecycle (TPLC) frameworks that can manage the iterative nature of AI software.
Implications for Patient Safety
If adopted, this competency-based assessment would represent a fundamental shift in regulatory philosophy, moving from the verification of technical specifications to the validation of clinical judgment. By treating AI more like a practitioner than a piece of equipment, the FDA aims to ensure these tools do not simply match data patterns but can be relied upon for cognitive reliability. For developers, this likely means a higher bar for entry, as they will need to prove their tools can navigate the nuances of real-world clinical scenarios.
The Path Forward
To manage the inherent uncertainty of generative AI during the premarket phase, the FDA is also exploring a greater reliance on postmarket monitoring. This would allow the agency to track the performance of devices in real-time after they have entered the clinic. It remains to be seen exactly how these benchmarks will be standardized or which clinical panels will be used to set the gold standard for AI competency.