One shared language for classifying any proposed AI use in pharmacy. Score every request against all five categories at the approval gate; most real uses touch more than one.
Dosing, interactions, substitutions, and any output that can reach a patient.
Control question: Who verifies the output before it touches a patient, and what happens when they disagree with it?
PHI exposure, training-data leakage, and access control.
Control question: Where does our data live, who can see it, and what may the vendor learn from it?
Hallucination, drift, and bias.
Control question: How is accuracy measured, against what baseline, and how often is it rechecked after go-live?
Automation errors and over-reliance.
Control question: What happens when staff stop checking, and how would we notice?
CMS, HIPAA, state board of pharmacy, and continuing-education integrity.
Control question: Which rules does this use touch, and who in our organization owns each one?