Abstract
Power can fail and be restored. A wrong clinical judgment stays with the patient. That is why the non-autonomy principle cannot remain an ethics statement. Inside the product structure, a physician must be able to confirm and to stop. Golden Clinical Loop is one cycle: AI proposes, a physician confirms, the action runs, and closing evidence is recorded.
This document does not claim autonomous clinical decisions, guaranteed accuracy, prevention of medical disputes, completed validation, regulatory clearance or clinical outcomes. Product screens use synthetic demo-account data and are not evidence of clinical validity or operating results.
- Document type
- Clinical safety principles
- Scope
- Clinical OS and Golden Clinical Loop
- Author
- AetherHeal
AetherHeal is building clinical execution infrastructure for hospitals and outpatient networks. Patient-access engines handle approved information, multilingual conversation, booking, and intake. Clinical-intelligence engines prepare encounter context, specialty inputs, and documentation for physician review. Care-execution engines coordinate permitted administrative actions, verification, ownership, and follow-through. These are not independent chatbots and Clinical OS is not a replacement central brain. Intelligence is distributed across bounded tasks; state, permissions, evidence, and accountability remain coordinated through a shared operating layer and a governed execution contract called the Golden Clinical Loop.
Physicians retain clinical judgment, treatment authority, correction, and final clinical responsibility. The existing hospital chart remains the system of record.
Execution authority × clinical impact
Execution authority and clinical impact are classified separately.
One axis distinguishes how far the system reads, proposes and changes state. The other distinguishes how far a task affects clinical judgment. The two axes cross to set the required controls.
01
Execution authority
- Read
- Queries approved information only and does not change the state of a reference system.
- Propose
- Suggests a draft or next action for review and does not execute.
- Write after approval
- Changes state only with a named person's approval and a permitted tool.
Clinical impact
- Operational
- Institutional work with no direct effect on clinical judgment.
- Within hospital protocol
- Handled only under an institution-approved protocol and a named clinical reviewer.
- Physician judgment
- Separated as work that needs regulatory review or a licensed physician's judgment.
This classification sets the boundary between internal execution controls and clinical impact. It is not automatically the same as legal or regulatory classification.
Per-cell control conditions, classification codes and the seven conditions for protocol-bound work: see the appendix PDF.
Responsibility and owner
Responsibility is split into seven named layers.
Clinical judgment, institutional protocol, system controls and deployment approval are never held by one party. Each responsibility has a different owner.
02
The unit of safety judgment is not a single answer. We review the record that continues until the task is complete in the patient's context: what inputs and uncertainty the proposal came from, who reviewed it, under what authority it ran, and what closing evidence was left in reference systems such as booking and the chart.
04
A higher level of claim needs new evidence and a new review procedure.
Product descriptions never exceed the scope of the evidence we currently hold.
05
Wording we use, and wording we do not.
Wording we use
- Physician-supervised clinical decision support
- Documentation completeness and risk-management support
- Reference, workflow, classification and measurement support
- Evidence-generation plans and regulatory review
Wording we do not use
- AI makes a patient's clinical decision on its own
- Guaranteed accuracy
- Prevents medical disputes
- Clinical validation or regulatory clearance is complete
- Improves clinical outcomes
- Replaces every EMR immediately
External references to read alongside: the WHO guidance on ethics and governance of AI for health [1], the FDA guidance on clinical decision support software [2], and the IMDRF document on clinical evaluation of software as a medical device [3].
06
A draft becomes a record only after physician review.
The medical record is the basis on which the appropriateness of care is judged afterwards. DockieTalkie Clinical Copilot organizes the draft record in a problem-oriented (POMR) structure, separating each problem's evidence, assessment, plan and gaps so the items for the physician to confirm are visible.
- 01Problem-oriented structure separating evidence, assessment and plan per problem
- 02A check of the statutory entries: chief complaint, diagnosis, course, treatment, and date and time
Electronic signature and record-integrity safeguards are being scoped against the electronic medical record requirements of Article 23 of the Medical Service Act [4]. Full compliance with legal e-signature and EMR retention and integrity requirements is a separate implementation scope.
Data boundaries (institutional boundary, no cross-institution reuse of original personal health information, processing purpose and processor role, scope of learning from corrections, consent and withdrawal, retention and deletion): see the appendix PDF.
07
References
- World Health Organization. Ethics and governance of artificial intelligence for health: WHO guidance. View the document
- U.S. Food and Drug Administration. Clinical Decision Support Software: Guidance for Industry and Food and Drug Administration Staff. View the document
- International Medical Device Regulators Forum. Software as a Medical Device: Clinical Evaluation. View the document
- Republic of Korea. Medical Service Act, Article 23 (Electronic Medical Records). National Law Information Center.
Full clinical safety document PDF →
Questions about the document's scope or citations: drjeeju@aetherheal.com