Governing autonomous AI in clinical care: the window is narrowing

Artificial intelligence systems in clinical care are moving beyond decision support into autonomous action-scheduling visits, initiating orders, and influencing insurance decisions such as prior authorization-yet, the governance infrastructure to deploy them safely is lagging. Primary care, where guideline-adherent care for a typical panel would require physicians to work more hours than there are in a day, is a natural deployment target and exemplifies this tension. While ample opportunities for artificial intelligence exist in almost every aspect of medicine, governance structures have not kept pace: accountability mechanisms, performance standards, and independent post-deployment monitoring remain weak, and the ability to oversee these systems often mirrors existing resource inequities. We propose that health systems and policymakers should risk-stratify autonomous actions by clinical consequence of error, tie procurement and value-based payment participation to minimum evaluation standards, and build shared governance infrastructure with sustainable funding so autonomous artificial intelligence narrows, rather than widens, existing care and equity gaps.