Medical AI Pipelines as Trust-Graphs: Accuracy is NOT Safety

Abstract

This record contains an audit-first structural map for medical AI deployment pipelines. The paper models end-to-end pipelines as trust-graphs and formalises the separation between (i) model correctness/performance and (ii) pipeline integrity in real-world clinical operations. Core structural claim (NO OVERCLAIM): any non-trivial medical AI pipeline has at least one responsibility binding layer R where trust is required and accountability must be assigned. This work is not a clinical claim and not a performance claim. Supplementary Information (SI) provides a minimal glossary, an R-candidate catalogue, and an audit record template. Axis: Peace / Non-violence / Dignity. Dual-use boundary: NO military, NO surveillance, NO coercion.

Ähnliche Arbeiten