Trust & Governance
Safety & Compliance
Built for clinical responsibility. Every design decision prioritises patient safety, clinician authority, and regulatory readiness.
Foundations
Core Safety Principles
Clinician-in-the-Loop
Every AI-generated output requires explicit clinician review and approval before it is saved, exported, or acted upon. The clinician is always the final decision-maker.
Mandatory Review Gates
No clinical note, referral, or prescription draft bypasses the review step. The system enforces approval workflows — there are no shortcuts.
Confidence Indicators
AI outputs include confidence scores and uncertainty flags. Areas with lower confidence are highlighted for clinician attention during review.
Complete Audit Trail
Every AI-generated draft, clinician edit, and approval action is logged with timestamps. Full traceability for compliance, quality, and clinical governance.
Access Controls
Role-based access ensures that only authorised clinicians can review, approve, and export clinical documentation. Administrative and clinical roles are separated.
Data Encryption
All health data is encrypted in transit (TLS 1.3) and at rest (AES-256). Audio recordings can be configured for automatic deletion after transcription.
1. Clinician-in-the-Loop Design
Miraa is architecturally designed so that no AI-generated content reaches a patient record, referral, or prescription without explicit clinician approval.
The clinician is not merely notified — they are the mandatory gatekeeper. The system cannot bypass this step. This is a deliberate design constraint, not an optional feature.
2. Mandatory Review of AI-Generated Outputs
Every type of AI-generated output is subject to mandatory clinician review. There are no exceptions, no bulk-approve shortcuts, and no automatic export paths.
This applies to all output types — from clinical notes to referral letters to prescription drafts. The system enforces individual review for each document before it can be saved or exported.
3. Audit Logs & Traceability
Every interaction with the system is recorded. AI outputs, clinician edits, approval actions, and exports are logged with timestamps and user identification.
- Timestamped AI output generation
- Record of all clinician edits and modifications
- Approval and rejection actions with clinician ID
- Export destinations and timestamps
- Confidence scores at time of generation
- Session recordings retention policy logging
Sample Audit Entry
4. Access Controls
Role-based access control ensures that only authorised clinicians can review, approve, and export clinical documentation. Administrative and clinical roles are separated.
Role-Based Access
Separate permissions for clinicians, practice managers, and administrators.
Authentication
Secure login with multi-factor authentication support for all users.
Session Management
Automatic session timeouts and audit logging for all access events.
5. Secure Handling of Sensitive Health Information
Health data is treated with the highest level of care. All data is encrypted in transit and at rest, and the system is designed to minimise data retention.
- TLS 1.3 encryption for all data in transit
- AES-256 encryption for all data at rest
- Stored in Supabase on AWS Australia (Sydney, ap-southeast-2)
- Supabase hosted platform is SOC 2 Type 2 compliant
- Configurable audio deletion after transcription
- No third-party data sharing without explicit consent
- Regular security audits and penetration testing
Encryption at Every Layer
Data protection is not an afterthought. It is embedded in every layer of the system architecture.
6. Privacy & Consent Workflow
Patient consent is built into the workflow. Clinics can configure their consent approach — per-visit verbal consent, written consent, or practice-level consent with opt-out. The system records consent status for every session.
Configurable Consent
Set consent policies per practice, per clinician, or per visit type.
Consent Recording
Consent status is logged and associated with every recording session.
Patient Opt-Out
Patients can opt out at any time. The system respects and records the preference.
7. Human Oversight for Prescriptions & Referrals
Prescription drafting and referral generation carry elevated clinical responsibility. These outputs receive additional safeguards beyond standard note review.
- Prescriptions are drafted only — never auto-submitted
- Clinician must verify drug, dose, frequency, and quantity
- Referral letters require explicit clinician sign-off
- Both output types are flagged for mandatory, individual review
- No batch-approve functionality for prescriptions or referrals
Drug: Amoxicillin 500mg
Dose: 1 capsule TDS
Duration: 5 days
Quantity: 15
Clinician Verification Required
This draft was pre-populated from consultation context. Verify all details against your clinical judgment before approving.
Not Autonomous Clinical Decision-Making
This distinction is fundamental to how Miraa is designed, built, and positioned.
Miraa is a clinical workflow copilot. It prepares context, captures consultation content, drafts notes and documents, and surfaces follow-up actions. It does not diagnose, treat, prescribe, or make clinical decisions.
The clinician is always the decision-maker. The system provides documentation support — the clinician provides clinical judgment. These roles are architecturally separated and cannot be conflated.
What the system does
- Transcribes clinical conversations
- Drafts structured clinical notes
- Pre-populates referral letters
- Suggests billing codes
- Captures follow-up actions
- Presents drafts for review
What the system does NOT do
- Diagnose medical conditions
- Make treatment decisions
- Prescribe medications autonomously
- Send referrals without approval
- Override clinician judgment
- Bypass the review step
9. Australia-First Compliance Posture
Miraa is designed and built for Australian healthcare from the ground up. Compliance is not retrofitted — it is foundational.
- Australian Privacy Principles (APPs) alignment
- My Health Records Act considerations
- Healthcare Identifiers Act awareness
- Australian data residency on AWS Sydney through Supabase
- Supabase SOC 2 Type 2 compliant platform controls
- Designed for TGA regulatory awareness (not a medical device)
- State and territory health records legislation awareness
10. International Readiness
While Australia is our primary market, the system architecture is designed for international compliance readiness.
- HIPAA-ready architecture for US market expansion
- GDPR-ready data handling for European markets
- Configurable data residency per region
- Modular compliance framework for new jurisdictions
Clear Boundaries
Being explicit about what the system does not do — and what should never be automated without review — is a core part of our safety commitment.
What the System Does Not Automate
- Clinical diagnosis or differential diagnosis
- Treatment planning or decision-making
- Medication prescribing or dispensing
- Referral sending (only drafting)
- Patient triage or risk assessment
- Billing submission (only code suggestions)
- Clinical coding validation
- Patient communication or advice
What Should Never Be Automated Without Review
- Prescription generation and submission
- Referral letters sent to other clinicians
- Clinical notes saved to patient records
- Billing claims submitted to Medicare
- Patient discharge instructions
- Pathology and imaging orders
- Specialist opinion letters
- Medico-legal documentation
Have Questions About Safety?
We welcome scrutiny. Join the waitlist for launch updates, or reach out directly at [email protected].
Join the Waitlist