Autonomous sterile compounding combines digital decision systems with a high-consequence physical process in which errors may affect product identity, dose, sterility, occupational exposure, workflow integrity, and medication release. Existing technologies can automate selected preparation, measurement, documentation, and verification tasks, but no single sensor, accuracy measure, artificial-intelligence model, or professional check is sufficient to establish system safety. This article proposes a cyber-physical safety architecture for autonomous sterile compounding cells. The architecture treats autonomy as conditional authority bounded by an authorized operating envelope, material and order controls, environmental and process sensing, independent verification, interlocks, hazard-specific safe states, accountable human supervision, evidence preservation, and controlled incident recovery. It distinguishes technical performance from medication-use decisions, automated execution from professional release authority, and successful implementation from validated benefit. The proposed relationships are organized as testable safety claims rather than assumed properties of automation. Evaluation would require component verification, aseptic and environmental qualification, fault injection, human-factors testing, workflow assessment, traceable safety-case evidence, and prospective monitoring under defined operating conditions. Changes in products, software, equipment, environment, staffing, or workflow would require reassessment of the applicable safety boundary. The architecture is conceptual and does not establish regulatory conformity, clinical effectiveness, universal applicability, or deployment readiness. Its original contribution is an integrated account of how physical containment, computational control, professional authority, and organizational governance may be connected without treating any individual component as a substitute for system-level safety assurance.