The American Medical Association has published a governance framework for agentic AI in healthcare, built on the principle that an AI agent must earn clinical authority through evidence before it performs clinical functions.
Agentic tools differ from earlier healthcare AI in one important way: they complete multi-step tasks autonomously instead of drafting output for a clinician to approve. That autonomy raises the stakes of an error, because a biased or wrong decision can reach a patient before any human notices.
The framework poses four questions, covering which tasks AI tools should be allowed to perform, in which situations those tools may act, and who is accountable when they fail. The AMA said the document reflects its position that AI systems delivering care in place of a physician must clear the same evidentiary bar expected of clinical interventions.
The guidance lands as health systems deploy AI agents for ambient documentation, diagnostics and patient monitoring, and as regulators and industry groups race to define oversight of tools that act on their own.
