Risk register
Known failure modes are managed openly.
Each engagement starts with a short register covering privacy exposure, wrong-output risk, staff override points, audit requirements, and operational fallback.
Safety & Compliance
AI is allowed into the workflow only after the boundaries are clear.
For NDIS, allied health, aged care, and education-adjacent work, the model is never the control plane. The workflow, risk register, audit trail, and human release authority are.
PII boundary
Identifying data has a designed path.
Names, participant identifiers, patient context, dates, and internal notes are isolated before model work wherever the workflow allows it.
Human release authority
Staff approve the final operational output.
AI can draft, classify, and queue. Claims, clinical notes, incident reports, and family communications still require human approval.
Audit trail
The system records what changed and why.
Outputs are designed to show source context, model involvement, reviewer action, and the final released version.
Operating model
The safety architecture is boring on purpose.
Most AI failures are not model failures. They are workflow failures: no owner, no fallback, no review state, no way to explain what happened.
Controlled intake
Structured fields, permission boundaries, and clear source records.
Constrained model role
The model handles only the narrow task it is trusted to handle.
Reviewed output
Human approval and logs before an output reaches a participant, patient, family, or portal.
Compliance posture
Mapped to the standards your operation already lives with.
NDIS
Practice Standards context
Incident, claims, plan-budget, record-keeping, and audit-prep workflows.AHPRA
Clinical accountability
Practitioner judgement stays visible; AI output is reviewable.Aged care
Quality and SIRS reporting
Structured incident capture and family communication controls.Privacy