The clinical and perioperative phases of medical work — the consultation, the operating theatre, the post-operative round, the structured handoff — have always rested on a single non-negotiable institutional principle: the practitioner of record remains the practitioner of record. The arrival of artificial intelligence at the point of clinical contact does not displace that principle. It either honours it by design or violates it by design. This paper proposes that the operational unit of safety engineering for an AI-mediated clinical assist is the adopt-or-override decision: every system suggestion is advisory, every interaction produces a recorded decision by the practitioner, and every recorded decision is bound to a cryptographically chained audit trail. It identifies four architectural commitments specific to the point of clinical contact, organised around two operational layers — practitioner-facing and institution-facing — and four consultation phases: anamnesis, examination, orientation, closure and report. One phase, orientation under the adopt-or-override grammar, is described in operational detail through four named components: evidence surfacing under permission constraints, suggestion-as-card, structured-note materialisation, and handoff letter generation. The paper proposes a six-point clinical-AI due-diligence procedure for medical directors and four evaluation axes: adoption discipline, structured-note completeness, audit reconstructability, and linguistic fidelity. The architecture inherits four foundational commitments from the series: source-grounded retrieval, permission-first governance, refusal as a first-class behaviour, and cryptographically chained audit trails. It operationalises Article 14 of the European Union Artificial Intelligence Act through the adopt-or-override grammar. Positioning under MDR Rule 11 is forward work; deployment outside investigational contexts is bounded by certification timing. No source code, deployment architecture, or proprietary implementation detail is disclosed. Part of a CLINETHIX preprint series on defendable clinical AI infrastructure, 2026. Related works: The Refusal Stack, doi:10.5281/zenodo.20257894; Sovereignty by Design, doi:10.5281/zenodo.20258141; Educator — Auditable Simulation as the Architectural Form of Clinical Training, CLINETHIX preprint series, 2026.
Fatima Azzahra MASTARI (Mon,) studied this question.