The November AI Accountability Turn: What the AMA–FSMB Report Changes for Physicians

Report, not a rulePublished 12 Nov 2026 · convened 14 Sep

A dated companion to Who Owns the Decision?. The original investigation and its earlier observations are unchanged; this report adds the November developments beside them.

What the sources establish

The American Medical Association published its report with the Federation of State Medical Boards on November 12, 2026, following a September 14 convening involving more than 200 participants. The publication is the new development; the meeting occurred earlier. [1]

The report records disagreement about oversight, professional licensing, device regulation and responsibility for harm. It does not establish a national AI medical license, transfer state-board authority to FSMB or adopt a binding rule governing every physician. [2]

An October 29 AMA liability brief argues for aligning responsibility with authority and the ability to address risk. [3]

The AMA’s November 10 account of newly adopted policy opposes physician licensure and physician titles for AI and describes advocacy concerning independent clinical judgment. Association policy and proposed legal protections do not themselves create an enacted defense for an individual physician. [4]

Original analysis

For a practicing physician, the useful next step is to translate the allocation of responsibility into observable facts about the care pathway. A broad declaration that a human remains responsible leaves unanswered which human controlled each relevant step and what that person could actually see or change.

Consider a synthetic example. A patient describes symptoms. Software drafts a note and suggests a diagnosis. The physician modifies the assessment. Another component generates patient instructions, and a later clinician encounters the resulting record. Identifying the final signer identifies one participant. It does not reconstruct every decision that shaped the information.

An investigation should locate the moment at which a suggestion acquired authority. Did it remain a draft? Did a clinician explicitly adopt it? Was it carried into another field automatically? Could a later reviewer see that it had been revised? These questions avoid assuming either that every generated statement is already a clinical decision or that a signature explains the entire sequence.

The proposed method is a control-and-evidence table for each function. Identify who selected the tool, who can change its settings, who sees its supporting information, who can reject or correct an output, who receives incident reports and what records survive a correction. Attach a policy, agreement, product document or observed test to each answer. Keep unresolved entries visibly unresolved.

The strongest institutional response deserves fair examination: the clinician retains final judgment and must review the draft. Test that response in the applicable workflow. Does the reviewer have sufficient information, time and authority to perform the promised review? If those conditions exist, document them. If they are inconsistent, identify who can repair them. A concern alone does not establish negligence or bad faith.

A developer may also point to training, product limitations and the distinction between drafting and acting. Those explanations should be checked against the function actually used. An investigation becomes more persuasive when it identifies a particular mismatch between a representation and an observed sequence, rather than treating every component in a connected system as interchangeable.

Patients need a clear explanation of what the technology does and who makes the clinical decisions. When an error reaches a record, an accurate correction and a clear care plan are immediate priorities. Preserve the relevant evidence so broader responsibility can be examined without delaying the correction itself.

This publication belongs beside “Who Owns the Decision?”, the DAX–Dragon–Epic investigation and UNFAITHFUL as a dated follow-up. Earlier observations should retain their dates and evidence. The next meaningful milestone is a documented change in policy, law, product behavior or clinical evidence. The November report provides a sharper set of questions; it does not establish that the accountability problem has been resolved.