FDA Official Believes GenAI-enabled Tools Will Need to Be Assessed Like a Physician

FDA Director Proposes GenAI Medical Devices Equivalent To Physician Evaluation Standards

During a panel discussion held Tuesday at The MedTech Conference organized by AdvaMed in Boston, FDA Director Rick Abdrorsson shared compelling insights regarding the future of artificial intelligence regulation.

Abramson, a Harvard‑trained radiologist and the director of the FDA’s Digital Health Center of Excellence, stated that generative AI tools may require evaluation methods similar to those applied to physicians during routine competency assessments—marking a significant shift from traditional approaches. Artificial intelligence is essentially advanced software, yet conventional regulatory frameworks had long treated such capabilities under different standards.

“We have spent two decades considering software as a medical device, accustomed to evaluating simple input‑output relationships,” Abramson explained. “In radiology, for example, feeding a chest X‑ray through an AI system yields a binary signal—’yes, no, pneumothorax.’ That straightforward outcome is conceptually easy to support.”

“Now we face a world where both input and output are virtually unlimited. This fundamentally changes how safety and efficacy can be assayed,” he continued. “What methods exist for providing credible evidentiary proof in such an unbounded environment?”

With GenAI, decision variables and outcomes become effectively infinite, demanding entirely new paradigms for scrutiny that differ from classic medical device validation.

“Essentially, we have moved into a scenario where evaluation spaces are unfenced. How do we address this? We simply lack established methodologies for generating that kind of evidence,” he declared.

Earlier this year, the FDA released a detailed discussion paper inviting industry stakeholders to weigh in on regulatory approaches for newly emerging GenAI‑enabled medical devices. Abramson outlined their thinking in detail.

“We concluded these technologies were beginning to behave somewhat like physicians rather than inert hardware such as heart valves or defibrillators,” Abramson noted. “Yet the process that would assess a doctor involves extensive exposure to varied clinical cases over time, evaluation of knowledge, judgment, communication skills, and adherence to safety constraints. No single physician receives daily testing across every conceivable diagnostic scenario.”

“The submitted feedback makes clear this approach is impractical. Evaluating every possible chest X‑ray throughout a career would be impossible,” he added.

The comment period concluded Friday, with the agency receiving over three hundred submissions. Moderator Geeta Pamidimukkala of AdvaMed posed a key question about next steps following public input.

“Directly answering, the administration intends to move forward with formal policy development. We will issue draft guidance—beyond an academic exercise or theoretical exploration. This initiative has received priority placement within the White House and at the HHS Secretary level.”

“When can we expect draft guidance? Abramson was deliberate in his timeline. He emphasized thorough review of the voluminous responses before proceeding, citing the importance of thoughtful synthesis rather than rapid iteration.”

He stressed the need to read and interpret comments attentively, stating the agency would not rely solely on automated analyses. “We intend to examine the feedback comprehensively and synthesize it deliberately. Given the significance, decisions should not emerge over mere weeks but certainly not span years,” he cautioned. “Draft policy guidance warrants an expected timeframe of months, followed by opportunities for further stakeholder input.”

The evolution of FDA thought processes—particularly the idea of assessing GenAI‑enabled devices alongside living human clinicians—has garnered attention. The American Medical Association responded through chief executive John Whyte, who authored a recent letter outlining recommendations for developing a robust regulatory framework. A notable concern raised by the AMA emphasizes practicality, influencing modifications in how the agency conceptualizes testing protocols for these emerging technologies.

Photo: steved_np3, Getty Images

Source link