It's a stack.
Every layer inherits the one beneath it.
It fails one layer at a time.
Downward toward the patient.
A scoping review published in PMC
found the same gap at every layer.
Everybody is governing one
and inheriting four.
Accountability is the layer nobody built.
Classical AI - The Rules Layer.
↳ Clinical protocols, eligibility logic,
↳ billing rules, compliance checks.
↳ If it fails the audit trail is clear.
Machine Learning - The Pattern Layer.
↳ Risk stratification, readmission prediction,
↳ diagnostic patterns, population screening.
↳ It inherits the rule nobody updated.
Deep Learning - The Imaging Layer.
↳ Radiology reads, pathology analysis,
↳ CT detection, MRI interpretation.
↳ It inherits the pattern nobody validated.
Generative AI - The Documentation Layer.
↳ AI scribes, discharge summaries,
↳ clinical notes, handoff documentation.
↳ It writes the unowned read into the chart.
Agentic AI - The Action Layer.
↳ Clinical memory, care planning,
↳ decision trace, autonomous action.
↳ It acts on a record carrying all four.
By the time the action layer moves,
four unnamed decisions are already inside the call.
Technical capability is outpacing governance.
Five layers. One framework.
The Governance Owner
names what each layer can do.
The Decision Owner holds
what reaches the bedside.
The Handoff is where
the inheritance stops.
The Accountability Gap™ (TAG™)
does not live in a layer.
It lives in what each layer passes down.
You already know which layer has no name on it.
That layer is not the problem.
It is what the four above it inherit.
Which layer would fail the named owner test today?
Mo Johnson, MD MBA is a cardiothoracic surgeon and the founder of GPe Research. Field Notes are short dispatches from the clinical AI accountability frontier, published alongside the MedicoVigilance™ newsletter at medicovigilance.org.
Follow the work on LinkedIn: linkedin.com/in/mo-johnson