The American Medical Association released a governance framework Oct. 13 for agentic AI that takes on clinical functions. The framework says an AI system’s clinical authority must be earned through evidence and maintained through ongoing performance.
The AMA said an AI system’s authority and oversight can change as its evidence, performance, risk or underlying technology changes. The framework centers on four questions:
1. Function: What clinical task is the AI performing? Tasks range from documentation and monitoring to triage, diagnosis, treatment and prescribing.
2. Evidence and risk: What shows the AI can perform that function, and what happens if it fails? Higher stakes and greater autonomy require stronger evidence.
3. Authority and oversight: What should the AI be allowed to do, from informing and recommending to deciding and acting, and under what conditions?
4. Accountability: Who is responsible for risks and outcomes? The AMA calls for assigning responsibility based on authority, control, knowledge of risk and ability to prevent harm. That responsibility could fall on developers, deployers, health systems, physicians, payers and others.
The AMA will use these questions to shape its work with physicians and other stakeholders on liability, licensure and regulation. The association clarified that “licensure” here means authorizing an AI system to operate in a defined clinical role, not a license to practice medicine. The framework also describes a continuous cycle in which AI is deployed, monitored, reassessed and refined based on what is learned.
The AMA’s Center for Digital Health and AI works with regulators, policymakers and technology leaders to set benchmarks for the safe, effective use of AI in medicine.
The post AMA: Agentic AI must earn clinical authority through evidence appeared first on Becker's Hospital Review | Healthcare News & Analysis.