Experts Propose Licensing Framework for Autonomous Clinical AI

A new regulatory model for autonomous medical AI

As artificial intelligence systems begin performing clinical tasks independently—taking patient histories, recommending treatments, and managing care—researchers at the University of Pennsylvania are proposing a regulatory framework that treats these tools like human clinicians rather than software.

In recent articles published in JAMA and JAMA Internal Medicine, LDI Senior Fellows Eric Bressman and Alon Bergman argue that autonomous clinical AI should be required to pass standardized competency exams, complete supervised deployment periods similar to medical residencies, and obtain time-limited licenses for defined scopes of practice.

The proposal comes as more than 250 AI-related bills move through state legislatures, creating what the researchers describe as a fragmented regulatory landscape. Utah recently passed legislation allowing AI to prescribe certain medications, while California prohibited using AI to deny insurance coverage.

The case for federal oversight

Bergman and co-authors Ezekiel Emanuel and Robert Wachter call for Congress to establish an Office of Clinical AI Oversight within the Department of Health and Human Services. This entity would certify autonomous clinical AI systems against performance standards before deployment.

Under the proposed framework, an AI system providing primary care would first need to score at or above the median of recent human test-takers on all three U.S. Medical Licensing Examination steps plus relevant board exams. The system would then enter a supervised deployment period, demonstrating adequate performance on a set patient volume with cases reviewed by human clinicians.

"Review is applied at the model layer, but clinical risk lives at the deployment layer," Bergman explained in an interview, distinguishing this approach from the White House's recent executive order focused on national security risks from frontier AI models.

Why it matters

The licensing framework addresses a critical gap as health systems deploy AI to address severe workforce shortages. Tens of thousands of physician vacancies cannot be filled through productivity gains alone—autonomous AI represents the only path to actually expanding clinical capacity. But without standardized competency requirements, patients in underserved areas could receive care from systems that haven't demonstrated baseline safety or effectiveness. A federal framework would prevent a patchwork of state standards while establishing clear pathways for developers.

Implementation challenges and next steps

The researchers acknowledge that their proposal requires new federal capabilities and Congressional action. Executive orders alone cannot preempt state law under the Supremacy Clause.

Bressman emphasized the urgency: "Federal inaction in this space has led to an emerging patchwork of state-level policies, and as long as the FDA and Congress stay silent, this fragmentation will only grow."

For health systems, the researchers recommend building AI governance structures immediately rather than waiting for federal frameworks. This includes establishing oversight functions for how tools are acquired and validated, conducting local performance validation on their own patient populations, and designing escalation and adverse event reporting processes.

The researchers also caution against overlooking harder-to-measure risks including deskilling of clinicians, automation bias, and what they term "model sycophancy"—when AI systems tell users what they want to hear rather than accurate clinical assessments.

Bergman advised state policymakers to focus on scope of practice, supervision, enforcement, and coverage rules rather than attempting to certify clinical competence independently. "The durable fix is a federal competency floor that state rules can sit atop," he said.

The framework details were first reported by the Leonard Davis Institute of Health Economics at the University of Pennsylvania, based on articles published in JAMA on April 29, 2026, and JAMA Internal Medicine in 2025.