FDA approval for AI isn’t working. What about medical licenses instead?

By Brittany Trang. Originally published in STAT News on November 19, 2025.

This article features work published by Carmel Shachar.

AI in health care is reaching a tipping point. Three things are happening at once.

• The FDA has approved over 1,000 medical devices that use AI, yet adoption remains low, at least by the data we have. 
• The FDA system of approving medical devices that have an AI or machine learning component does not encompass the full spectrum of AI in clinical settings, especially the generative AI applications that show the most transformative promise.
• At the same time, “There’s pressure to get this stuff into people’s hands,” said Maulin Shah, chief medical information officer at the Providence health system. “There’s pressure to improve patient care, to reduce burden, to see the value of AI as quickly as we can, which is why we’re all struggling so hard with this.”

Enter a new proposal from four health policy experts: What if instead of the FDA approving AI models through the “software as a medical device” (SaMD) framework, we used a different and more flexible framework that we already have: licensing AI as a medical practitioner? 

Under the proposed framework, AI developers would get a license for the base model, and the licensing authority would specify the AI models’ scope of practice. The AI would need to meet minimum performance thresholds in those areas, undergo “a period of supervised training and practice in an accredited clinical environment,” and have multiple surveillance mechanisms afterward. Those might include regular retesting and a discipline board for complaints. 

Read the full article.