Cad Care RISE Resilience and Inclusive Solutions for Emergencies
Pre-launchCad Care RISE is in development. Public launch is expected in 2028, with air ambulance capability targeted for 2030. Nothing on this site describes a service that is currently operating.
CMP-04 · AI GOVERNANCE

AI governance in RISE

Which parts of the platform are artificial intelligence, exactly what they are permitted to do, and where the human sits in every loop.

Principle

The boundary, made concrete

The design philosophy is uncompromising: the AI assists, humans decide. Every AI component is bounded to a narrow mandate, every output is explainable and logged, and the entire layer can be disabled with the platform continuing to operate as a deterministic system.

That last property is the one that matters under scrutiny. A platform whose AI cannot be switched off without the platform failing has not implemented human oversight; it has implemented a dependency.

Orchestration
A backend orchestrator decides when a model runs, assembles its grounded context, and routes its output to a human queue — never directly to a clinical or dispatch decision.
Grounded reasoning
Models reason over retrieved, approved reference material rather than free recall, which keeps outputs grounded and auditable.
Guardrails
Structured outputs, validation of every response, deterministic fallback, and a global kill-switch.
Auditability
Each AI action is written to an immutable log with its inputs, outputs and rationale, and is reconstructible by incident.
Human-in-the-loop enforcement
Enforced in the orchestrator, not left to good intentions or to training. A configuration that removes human review is not available to deploying organisations.
Hard limits

What the AI is never permitted to do

  • No aircraft is launched by the platform. Dispatch is recommended by the system and committed by a named authoriser, with the reason recorded.
  • No diagnosis is made. Triage support suggests an order; a clinician confirms it.
  • No aid request is declined by a model. Verification decisions that refuse a request are made by a person, and carry an appeal route.
  • No consent is inferred. Emergency access paths exist and are recorded; they are never granted by a model's judgement of clinical necessity.
  • No resource is committed autonomously. Predictive output shapes standby posture and pre-positioning; commitment is a human act.
  • No model output is presented without its uncertainty. A prediction offered without a confidence range invites false precision.
Who is accountable

Oversight function

SAHI recommends dedicated AI units or nodal cells within health departments and institutions, to lead AI strategy, use-case prioritisation, tool assessment, deployment oversight and lifecycle management.

The programme maintains an AI oversight function with a named accountable owner, an agent audit log reviewable by compliance, per-component enable and disable controls, and a change-audit note on every configuration change. Deploying organisations are supported in establishing the equivalent function on their own side, because oversight that exists only at the vendor is not oversight.