Medication decisions supported by artificial intelligence are often described as if responsibility remains wholly with the pharmacist because a human formally approves the final action. This view overlooks how data selection, model design, interface presentation, organizational configuration, professional judgment, and follow-up jointly produce the decision. This article develops a proposed Responsibility-Allocation Model for co-produced medication decisions. The model treats each decision as a responsibility-bearing episode extending across data, recommendation, approval, and follow-up. At every stage, four dimensions are examined separately: contribution, meaning the input supplied to the decision; control, meaning the feasible capacity to inspect or alter the process; authority, meaning legitimate power to approve, reject, or escalate; and liability, meaning potential legal or institutional answerability. The model maps these dimensions across the pharmacist, algorithm developer or vendor, and deploying organization. It also introduces proposed rules for non-equivalence among responsibility dimensions, effective professional control, authority–control alignment, documented disagreement, and reassessment after material change. Evaluation requires evidence of technical validity, joint human–algorithm task performance, workflow effects, medication safety, equity, traceability, organizational response, and longitudinal stability. The model does not determine negligence, allocate legal liability, establish clinical benefit, or authorize deployment. Its original contribution is a structured vocabulary and testable governance architecture for examining responsibility without reducing it either to algorithmic influence or to the pharmacist’s final click, enabling later empirical testing across pharmacy tasks and organizational contexts.