FDA Weighs Doctor-Style Testing for AI Medical Devices

A New Way to Judge Machine Judgment

The Food and Drug Administration is exploring a fundamentally different way to evaluate artificial intelligence in medicine: judging software the way it would judge a physician. In a discussion paper released this week, the agency floated a "competency-based approach" for assessing generative AI-enabled medical devices, testing their reasoning and decision-making in a manner reminiscent of how doctors are certified and evaluated for clinical competency 15. The paper is not a formal rule but a request for public feedback, reflecting the FDA's acknowledgment that its existing device-review framework, built largely around static, single-function software, is straining to keep pace with generative and adaptive AI systems that can reason, generate text, and adjust behavior in ways traditional medical devices never did 2.

Why Existing Rules Are Under Pressure

Regulators face a genuinely novel problem set. Generative AI tools do not behave like fixed algorithms; their outputs can vary, evolve, or be difficult to fully explain, which complicates the FDA's traditional premarket review model built around predictable, reproducible performance 5. The scale of the challenge is growing quickly: more than 1,357 AI-enabled medical devices have now received FDA authorization, roughly double the number cleared through 2022, according to reporting that also flagged troubling real-world failures, including AI systems misidentifying body parts and introducing new risks during surgical procedures 7. That surge in both approvals and incidents underscores why the agency is rethinking oversight rather than simply extending old rules to new technology.

Breakthroughs and Growing Scrutiny

At the same time, the FDA has continued to clear high-profile AI tools through its existing breakthrough device pathways, including PanCancer Detect, a tool aimed at early cancer detection, illustrating that the agency is simultaneously encouraging innovation while tightening scrutiny of how these tools are validated and monitored after approval 36. This dual posture — welcoming advanced diagnostic AI while demanding more rigorous evidence of safety and effectiveness — appears consistent across the FDA's recent breakthrough-device activity.

Imaging Gains, Security Gaps

Medical imaging remains one of the clearest success stories for healthcare AI, with algorithms improving diagnostic speed and accuracy and contributing to a growing share of FDA clearances in radiology 8. Yet the same connectivity and complexity that make these tools powerful also make them vulnerable. Cybersecurity experts warn that AI-enabled devices, if not built with security as a foundational design principle, represent a high-stakes gamble, since compromised imaging or diagnostic systems could directly endanger patients 4.

What Comes Next

Taken together, the coverage points to a regulatory system in transition: the FDA is testing new evaluative frameworks, approving cutting-edge diagnostic tools, confronting documented safety failures, and fielding warnings about cybersecurity risk — all at once. The competency-based proposal signals that regulators increasingly view sophisticated medical AI less like conventional software and more like an autonomous clinical actor whose judgment, not just its code, must be tested.