A Licence for AI Doctors?

Companies developing agentic AI for healthcare are reportedly discussing a national licensing framework for AI clinicians in the United States.

According to Inside Health Policy, the proposed national route would avoid requiring autonomous clinical systems to navigate the existing state-by-state physician-licensing structure, including the Interstate Medical Licensure Compact.

This represents a remarkable shift. Until recently, the dominant questions were whether AI could diagnose, reason clinically or operate safely without continuous physician supervision.

The emerging question is much more concrete:

What would it mean to license an AI clinician?

A credible framework would potentially need to define:

• Competency requirements;
• Permitted scope of practice;
• Supervision, abstention and escalation rules;
• Geographic authority to operate;
• Continuing monitoring and recertification;
• Accountability when the system causes harm.

My takeaway

This is where the autonomous-clinical-AI discussion was heading.

If AI begins functioning as a genuine clinical actor rather than merely as a tool, medicine may eventually need something analogous to qualification, competency testing, certification, scope of practice, continuing evaluation—and perhaps licensure.

But licensing immediately exposes an even harder question: liability.

If an AI can effectively hold a licence, who carries the malpractice risk—the developer, deploying health system, supervising clinician or another accountable institution?

The report says informed-consent changes are also being discussed, including a proposal that could shift some liability toward patients. That possibility deserves particularly careful scrutiny. Consent should inform patients and support meaningful choice; it should not become a mechanism for transferring poorly controlled technological risk to them.

There is also substantial institutional resistance to treating AI as a licensed practitioner. The American Medical Association recently supported a proposed Washington Medical Commission interpretation that, under current law, AI cannot receive a licence to practise medicine or independently diagnose, treat or prescribe.

Signal, not law

No national AI-clinician licence currently exists. These are reported industry-policy discussions, not enacted legislation or a settled regulatory proposal.

Nevertheless, they reveal the direction of travel:

AI as a tool → AI as a teammate → AI as a clinical actor → AI as a regulated professional category?